Predictive Modeling in Biomedical Data Mining

and Analysis 1st Edition- eBook PDF pdf download

https://ebooksecure.com/download/predictive-modeling-in-
biomedical-data-mining-and-analysis-ebook-pdf/

Download more ebook from https://ebooksecure.com

We believe these products will be a great fit for you. Click
the link to download now, or visit ebooksecure.com
to discover even more!

(eBook PDF) Data Mining and Predictive Analytics 2nd

Edition

http://ebooksecure.com/product/ebook-pdf-data-mining-and-
predictive-analytics-2nd-edition/

(eBook PDF) Handbook of Statistical Analysis and Data

Mining Applications 2nd Edition

http://ebooksecure.com/product/ebook-pdf-handbook-of-statistical-
analysis-and-data-mining-applications-2nd-edition/

Handbook of statistical analysis and data mining

applications Second Edition Elder - eBook PDF

https://ebooksecure.com/download/handbook-of-statistical-
analysis-and-data-mining-applications-ebook-pdf/

(eBook PDF) Statistics, Data Analysis, and Decision

Modeling 5th Edition

http://ebooksecure.com/product/ebook-pdf-statistics-data-
analysis-and-decision-modeling-5th-edition/
(eBook PDF) Introduction to Business Data Mining 1st
Edition

http://ebooksecure.com/product/ebook-pdf-introduction-to-
business-data-mining-1st-edition/

Quantitative Analysis and Modeling of Earth and

Environmental Data: Space-Time and Spacetime Data
Considerations 1st Edition Jiaping Wu - eBook PDF

https://ebooksecure.com/download/quantitative-analysis-and-
modeling-of-earth-and-environmental-data-space-time-and-
spacetime-data-considerations-ebook-pdf/

(eBook PDF) Applied Predictive Modeling 2nd 2018

Edition

http://ebooksecure.com/product/ebook-pdf-applied-predictive-
modeling-2nd-2018-edition/

(eBook PDF) Data Mining Concepts and Techniques 3rd

http://ebooksecure.com/product/ebook-pdf-data-mining-concepts-
and-techniques-3rd/

Big Data Mining for Climate Change 1st edition - eBook

PDF

https://ebooksecure.com/download/big-data-mining-for-climate-
change-ebook-pdf/
 Predictive Modeling
in Biomedical Data Mining
and Analysis
This page intentionally left blank
 Predictive Modeling
in Biomedical Data Mining
and Analysis

Edited by
Sudipta Roy
Department of Artificial Intelligence and Data Science, Jio Institute,
Navi Mumbai, Maharashtra, India

Lalit Mohan Goyal

Department of Computer Engineering, J C Bose University of Science
and Technology, YMCA, Faridabad, India

Valentina E. Balas
Professor of Automation and Applied Informatics, Aurel Vlaicu University
of Arad, Arad, Romania

Basant Agarwal
Department of Computer Science and Engineering, Indian Institute
of Information Technology Kota, Jaipur, Rajasthan, India

Mamta Mittal
Delhi Skill and Entrepreneurship University, New Delhi, India
Academic Press is an imprint of Elsevier
125 London Wall, London EC2Y 5AS, United Kingdom
525 B Street, Suite 1650, San Diego, CA 92101, United States
50 Hampshire Street, 5th Floor, Cambridge, MA 02139, United States
The Boulevard, Langford Lane, Kidlington, Oxford OX5 1GB, United Kingdom

Copyright © 2022 Elsevier Inc. All rights reserved.

No part of this publication may be reproduced or transmitted in any form or by any means, electronic or
mechanical, including photocopying, recording, or any information storage and retrieval system, without
permission in writing from the publisher. Details on how to seek permission, further information about the
Publisher’s permissions policies and our arrangements with organizations such as the Copyright Clearance
Center and the Copyright Licensing Agency, can be found at our website: www.elsevier.com/permissions.

This book and the individual contributions contained in it are protected under copyright by the Publisher
(other than as may be noted herein).
Notices
Knowledge and best practice in this field are constantly changing. As new research and experience broaden our
understanding, changes in research methods, professional practices, or medical treatment may become
necessary.

Practitioners and researchers must always rely on their own experience and knowledge in evaluating and using
any information, methods, compounds, or experiments described herein. In using such information or methods
they should be mindful of their own safety and the safety of others, including parties for whom they have a
professional responsibility.

To the fullest extent of the law, neither the Publisher nor the authors, contributors, or editors, assume any liability
for any injury and/or damage to persons or property as a matter of products liability, negligence or otherwise, or
from any use or operation of any methods, products, instructions, or ideas contained in the material herein.

ISBN: 978-0-323-99864-2

For information on all Academic Press publications

visit our website at https://www.elsevier.com/books-and-journals

Publisher: Mara E. Conner

Acquisitions Editor: Chris Katsaropoulos
Editorial Project Manager: Fernanda A. Oliveira
Production Project Manager: Anitha Sivaraj
Cover Designer: Greg Harris

Typeset by STRAIVE, India

Contents

Contributors xi
About the Editors xv
Preface xix

1. Data mining with deep learning in biomedical data 1

Kuldeep Singh and Jyoteesh Malhotra

1. Introduction 1
2. Role of deep learning techniques in epileptic seizure
detection 3
3. Proposed method of seizure detection 5
4. Results and discussion 12
5. Conclusions 16
References 16

2. Applications of supervised machine learning techniques

with the goal of medical analysis and prediction:
A case study of breast cancer 21
Koushal Kumar and Bhagwati Prasad Pande

1. Introduction 21
2. A brief literature survey 23
3. Dataset and modus operandi 24
4. Data visualization 30
5. Feature selection and dimensionality reduction 33
6. Experimental results and discussions 39
7. Conclusions 45
References 46

v
vi Contents

3. Medical decision support system using data mining 49

N.L. Taranath, H.R. Roopashree, A.C. Yogeesh, L.M. Darshan, and
C.K. Subbaraya

1. Introduction 49
2. Medical decision support system: A review 50
3. Ontological representation of MDSS 53
4. Integrated medical decision support system 57
5. Conclusion and future enhancement 62
References 63

4. Role of AI techniques in enhancing multi-modality

medical image fusion results 65
Harmeet Kaur and Satish Kumar

1. Introduction 65
2. Modalities 66
3. Fusion process 67
4. AI based fusion 70
5. Evaluation 73
6. Experimental results 75
7. Conclusion and future scope 79
Acknowledgment 79
References 79

5. A comparative performance analysis of backpropagation

training optimizers to estimate clinical gait mechanics 83
Jyotindra Narayan, Sanchit Jhunjhunwala, Shivansh Mishra,
and Santosha K. Dwivedy

1. Introduction 83
2. Methods: Related work and dataset 86
3. Backpropagation neural network and training optimizers 88
4. BPNN implementation 92
 Contents vii

5. Results and discussions 94

6. Conclusions 101
References 102

6. High-performance medicine in cognitive impairment:

Brain–computer interfacing for prodromal
Alzheimer’s disease 105
H.M.K.K.M.B. Herath, R.G.D. Dhanushi, and B.G.D.A. Madhusanka

1. Introduction 105
2. Related works 108
3. Methodology 109
4. Results 115
5. Conclusion 119
References 120

7. Brain tumor classifications by gradient and XG boosting

machine learning models 123
Nalini Chintalapudi, Gopi Battineni, Lalit Mohan Goyal,
and Francesco Amenta

1. Introduction 123
2. Research background 125
3. Methods 126
4. Results and discussions 132
5. Conclusions 135
Conflicts of interest 135
References 135

8. Biofeedback method for human–computer interaction

to improve elder caring: Eye-gaze tracking 137
B.G.D.A. Madhusanka, Sureswaran Ramadass, Premkumar Rajagopal,
and H.M.K.K.M.B. Herath

1. Introduction 137
viii Contents

2. Anatomy of the human eye 138

3. Overview of eye-gaze tracking 140
4. Eye-gaze tracking for human–computer interaction 142
5. Proposed design 143
6. Results 147
7. Conclusion 151
References 152

9. Prediction of blood screening parameters for preliminary

analysis using neural networks 157
Aman Kataria, Divya Agrawal, Sita Rani, Vinod Karar,
and Meetali Chauhan

1. Introduction 157
2. Related work 158
3. Methodology 160
4. Results 163
5. Conclusion 167
References 167

10. Classification of hypertension using an

improved unsupervised learning technique and
image processing 171
Usharani Bhimavarapu and Mamta Mittal

1. Introduction 171
2. Related work 174
3. Methodology 175
4. Experimental results 178
5. Conclusion 184
References 184
 Contents ix

11. Biomedical data visualization and clinical decision-making

in rodents using a multi-usage wireless brain stimulator
with a novel embedded design 187
V. Milner Paul, Loitongbam Surajkumar Singh, S.R. Boselin Prabhu,
T. Jarin, Shuma Adhikari, and S. Sophia

1. Introduction 187
2. Architectural design and circuit modeling 189
3. Implementation and experimental verification 193
4. Results and discussions 201
5. Conclusion and future directions 202
References 204

12. LSTM neural network-based classification of sensory

signals for healthy and unhealthy gait assessment 207
Jyotindra Narayan, Sanghamitra Johri, and Santosha K. Dwivedy

1. Introduction 207
2. Dataset collection 209
3. LSTM neural network model 209
4. Implementation of LSTM neural network 215
5. Results and discussions 217
6. Conclusions 221
References 221

13. Data-driven machine learning: A new approach to

process and utilize biomedical data 225
Kalpana, Aditya Srivastava, and Shashank Jha

1. An introduction to artificial intelligence and machine

learning in healthcare 225
2. Challenges and roadblocks to be addressed 231
3. The need to address these issues 238
4. Recommendations and guidelines for the improvement
of ML-based algorithms 238
x Contents

5. Applications in the present scenarios 241

6. Future prospects and conclusion 244
References 246

14. Multiobjective evolutionary algorithm based on

decomposition for feature selection in medical diagnosis 253
Sudhansu Shekhar Patra, Mamta Mittal, and Om Prakash Jena
1. Introduction 253
2. Medical applications 255
3. Feature selection 257
4. Literature review 261
5. Metaheuristics and MOO 262
6. Multiobjective optimization problems (MOOPs) 266
7. Role of EA in MOO 272
8. MOEA based on decomposition 274
9. Application of MOEA/D in feature selection for
medical diagnosis 281
10. Experimental results 286
11. Conclusion 289
References 289

15. Machine learning techniques in healthcare informatics:

Showcasing prediction of type 2 diabetes mellitus
disease using lifestyle data 295
Majid Bashir Malik, Shahid Mohammad Ganie, and Tasleem Arif

1. Introduction 295
2. Machine learning in healthcare 296
3. Proposed framework 302
4. Results and discussion 303
5. Conclusion and future scope 306
References 309

Index 313
Contributors

Shuma Adhikari Department of Electrical Engineering, National Institute of Technology

Manipur (NITM), Imphal, India
Divya Agrawal CSIR-CSIO, Chandigarh, India
Francesco Amenta Clinical Research Centre, School of Medicinal and Health Products
Sciences, University of Camerino, Camerino; Research Department, International Radio
Medical Centre (C.I.R.M.), Rome, Italy
Tasleem Arif Department of Information Technology, BGSB University, Rajouri, UT J&K,
India
Gopi Battineni Clinical Research Centre, School of Medicinal and Health Products
Sciences, University of Camerino, Camerino, Italy
Usharani Bhimavarapu Department of Computer Science and Engineering, Koneru
Lakshmaiah Education Foundation, Vaddeswaram, Andhra Pradesh, India
Meetali Chauhan Department of Computer Science and Engineering, Gulzar Institute of
Engineering and Technology (Affiliated To I.K.G. Punjab Technical University, Kapurthala)
Gulzar Group of Institutes, Khanna (Ludhiana), Punjab, India
Nalini Chintalapudi Clinical Research Centre, School of Medicinal and Health Products
Sciences, University of Camerino, Camerino, Italy
L.M. Darshan School of CSE, REVA University, Bengaluru, Karnataka, India
R.G.D. Dhanushi Faculty of Livestock Fisheries & Nutrition, Wayamba University of Sri
Lanka, Makandura, Sri Lanka
Santosha K. Dwivedy Department of Mechanical Engineering, Indian Institute of
Technology Guwahati, Guwahati, Assam, India
Shahid Mohammad Ganie Department of Computer Sciences, BGSB University, Rajouri,
UT J&K, India
Lalit Mohan Goyal Department of Computer Engineering, J C Bose University of Science
and Technology, YMCA, Faridabad, India
H.M.K.K.M.B. Herath Faculty of Computing and IT, Sri Lanka Technological Campus,
Padukka, Sri Lanka

xi
xii Contributors

T. Jarin Department of Electrical and Electronics Engineering, Jyothi Engineering College,

Thrissur, India
Om Prakash Jena Department of Computer Science, Ravenshaw University, Cuttack,
India
Shashank Jha Department of Biotechnology, Dr. Ambedkar Institute of Technology for
Handicapped, Kanpur, Uttar Pradesh, India
Sanchit Jhunjhunwala Department of Mechanical Engineering, Indian Institute of
Technology Guwahati, Guwahati, Assam, India
Sanghamitra Johri Department of Mechanical Engineering, Indian Institute of
Technology Guwahati, Guwahati, Assam, India
Kalpana Department of Biotechnology, Dr. Ambedkar Institute of Technology for
Handicapped, Kanpur, Uttar Pradesh, India
Vinod Karar CSIR-CSIO, Chandigarh, India
Aman Kataria CSIR-CSIO, Chandigarh, India
Harmeet Kaur DCSA, Panjab University, Chandigarh, India
Koushal Kumar Sikh National College, Qadian, Guru Nanak Dev University, Amritsar,
Punjab, India
Satish Kumar SSG Regional Centre Hoshiarpur, Panjab University, Chandigarh, India
B.G.D.A. Madhusanka School of Science and Engineering, Malaysia University of Science
and Technology (MUST), Petaling Jaya, Malaysia
Jyoteesh Malhotra Department of Engineering and Technology, Guru Nanak Dev
University Regional Campus Jalandhar, Punjab, India
Majid Bashir Malik Department of Computer Sciences, BGSB University, Rajouri, UT J&K,
India
V. Milner Paul Department of Electrical Engineering, National Institute of Technology
Manipur (NITM), Imphal, India
Shivansh Mishra Department of Mechanical Engineering, Sardar Vallabhbhai National
Institute of Technology, Surat, India
Mamta Mittal Delhi Skill and Entrepreneurship University, New Delhi, India
Jyotindra Narayan Department of Mechanical Engineering, Indian Institute of
Technology Guwahati, Guwahati, Assam, India
Bhagwati Prasad Pande Department of Computer Applications, LSM Government PG
College, Pithoragarh, Uttarakhand, India
 Contributors xiii

Sudhansu Shekhar Patra School of Computer Applications, KIIT Deemed to be

University, Bhubaneswar, India
S.R. Boselin Prabhu Department of Electronics and Communication Engineering, Surya
Engineering College, Mettukadai, India
Premkumar Rajagopal Malaysia University of Science and Technology (MUST), Petaling
Jaya, Malaysia
Sureswaran Ramadass School of Science and Engineering, Malaysia University of Science
and Technology (MUST), Petaling Jaya, Malaysia
Sita Rani Department of Computer Science and Engineering, Gulzar Institute of
Engineering and Technology (Affiliated To I.K.G. Punjab Technical University,
Kapurthala) Gulzar Group of Institutes, Khanna (Ludhiana), Punjab, India
H.R. Roopashree Department of CS & E, GSSSITEW, Mysuru, Karnataka, India
Kuldeep Singh Department of Electronics Technology, Guru Nanak Dev University
Amritsar, Punjab, India
Loitongbam Surajkumar Singh Department of Electronics & Communication
Engineering, National Institute of Technology Manipur (NITM), Imphal, India
S. Sophia Department of Electronics and Communication Engineering, Sri Krishna
College of Engineering and Technology, Kuniyamuthur, India
Aditya Srivastava Department of Biomedical Engineering, Indian Institute of Technology
Hyderabad, Sangareddy, Kandi, Telangana, India
C.K. Subbaraya Adichuchanagiri University, Nagamangala, Karnataka, India
N.L. Taranath Department of CS & E, Graphic Era Hill University, Dehradun, Uttarakhand,
India
A.C. Yogeesh Department of CS & E, Government Engineering College, Kushalnagar,
Karnataka, India
This page intentionally left blank
About the Editors

Dr. Sudipta Roy is working as an assistant professor in the Artificial Intelligence and Data
Science Department at JIO Institute, Navi Mumbai, Maharashtra, India. Prior to that, he
was a postdoctoral research associate at Washington University in St. Louis, MO, United
States. He has received his PhD in Computer Science and Engineering from the Depart-
ment of Computer Science and Engineering, University of Calcutta, Kolkata, West Bengal,
India. He is the author of more than 50 publications in refereed international journals and
conference proceedings published by IEEE, Springer, Elsevier, and many other publishers.
He has authored/edited four books and many book chapters. He holds a US patent in
medical image processing and has filed an Indian patent in the field of smart agricultural
systems. He has served as a regular reviewer for many international journals including
those published by IEEE, Springer, Elsevier, IET, and many other publishers, and interna-
tional conferences. He has served as an international advisory committee member and
program committee member of INDIAcom-2020, AICAE-2019, INDIACom-2019, CAAI
2018, ICAITA-2018, ICSESS-2018, INDIACom-2018, ISICO-2017, AICE-2017, and many
other conferences. Currently, he is serving as associate editor of IEEE Access (IEEE) and
International Journal of Computer Vision and Image Processing (IJCVIP; IGI Global) and
topic editor of Journal of Imaging (MDPI). In recognition of his exceptional contributions
to the IEEE Access journal as associate editor, the IEEE Access Editorial Board and Editorial
Office honored him as an Outstanding Associate Editor of 2020. He has more than 5 years
of experience in teaching and research. His fields of research interests are biomedical
image analysis, image processing, steganography, artificial intelligence, big data analysis,
machine learning, and big data technologies.

Dr. Lalit Mohan Goyal has completed PhD in Computer Engineering from Jamia Millia
Islamia, New Delhi, India, MTech (Honors) in Information Technology from Guru Gobind
Singh Indraprastha University, New Delhi, India, and BTech (Honors) in Computer Engi-
neering from Kurukshetra University, Kurukshetra, India. He has 17 years of teaching
experience in the areas of theory of computation, parallel and random algorithms, distrib-
uted data mining, and cloud computing. He has completed a project sponsored by the
Indian Council of Medical Research, Delhi. He has published and communicated more
than 40 research papers in SCI, SCIE, and Scopus-indexed journals and attended many
workshops, Faculty Development Programs, and seminars. He has filed nine patents in
the area of artificial intelligence and deep learning, out of which four have been granted
and others are published online. He is a reviewer of many reputed journals and confer-
ences. He is a series editor for CRC Press, Taylor & Francis, and has edited many books

xv
xvi About the Editors

for Elsevier and Springer. Presently, he is working in the Department of Computer Engi-
neering, J.C. Bose University of Science and Technology, YMCA, Faridabad, India.

Prof. Valentina E. Balas is currently full professor in the Department of Automatics and
Applied Software at the Faculty of Engineering, Aurel Vlaicu University of Arad, Romania.
She holds a PhD cum laude in Applied Electronics and Telecommunications from Poly-
technic University of Timisoara. Dr. Balas is the author of more than 400 research papers
in refereed journals and international conferences. Her research interests are in intelligent
systems, fuzzy control, soft computing, smart sensors, information fusion, and modeling
and simulation. She is the editor-in-chief of the International Journal of Advanced Intel-
ligence Paradigms (IJAIP) and International Journal of Computational Systems Engineer-
ing (IJCSysE), editorial board member of several national and international journals, and
expert evaluator for national/international projects and PhD theses. Dr. Balas is the direc-
tor of Intelligent Systems Research Centre in Aurel Vlaicu University of Arad and director
of the Department of International Relations, Programs and Projects in the same univer-
sity. She served as the general chair for nine editions of the International Workshop on Soft
Computing and Applications (SOFA) organized during the period 2005–2020 and held in
Romania and Hungary. Dr. Balas participated in many international conferences as orga-
nizer; honorary chair; session chair; member of the steering, advisory, or international
program committees; and keynote speaker. Recently, she was working on a national pro-
ject with EU funding support “BioCell-NanoART ¼ Novel Bio-inspired Cellular Nano-
Architectures—For Digital Integrated Circuits,” 3M Euro from the National Authority
for Scientific Research and Innovation. She is a member of the European Society for Fuzzy
Logic and Technology (EUSFLAT), member of the Society for Industrial and Applied
Mathematics (SIAM), a senior member of IEEE, member of Technical Committee—Fuzzy
Systems (IEEE Computational Intelligence Society), chair of the Task Force 14 in Technical
Committee—Emergent Technologies (IEEE CIS), and member of Technical Committee—
Soft Computing (IEEE SMCS). Dr. Balas was past vice president of the International Fuzzy
Systems Association (IFSA) Council (2013–2015), is a joint secretary of the governing
council of the Forum for Interdisciplinary Mathematics (FIM, a multidisciplinary aca-
demic body based in India), and is a recipient of the “Tudor Tanasescu” Prize from the
Romanian Academy for contributions in the field of soft computing methods (2019).

Dr. Basant Agarwal is working as an assistant professor at the Indian Institute of Informa-
tion Technology Kota (IIIT-Kota), India, which is an institute of national importance. He
holds a PhD and MTech from the Department of Computer Science and Engineering,
Malaviya National Institute of Technology Jaipur, India. He has more than nine years of
experience in research and teaching. He has worked as a postdoc research fellow at the
Norwegian University of Science and Technology (NTNU), Norway, under the prestigious
European Research Consortium for Informatics and Mathematics (ERCIM) fellowship in
2016. He has also worked as a research scientist at Temasek Laboratories, National
 About the Editors xvii

University of Singapore (NUS), Singapore. His research interests are in artificial intelli-
gence, cyber-physical systems, text mining, natural language processing, machine learn-
ing, deep learning, intelligent systems, expert systems, and related areas.

Dr. Mamta Mittal is working as program head and associate professor (Data Analytics and
Data Science) in Delhi Skill and Entrepreneurship University (under the Government of
NCT Delhi), New Delhi, India. She received her PhD in Computer Science and Engineering
from Thapar University, Patiala; MTech (Honors) in Computer Science and Engineering
from YMCA, Faridabad; and BTech in Computer Science and Engineering from Kurukshe-
tra University, Kurukshetra, in 2001. She has been teaching for the last 18 years and spe-
cializes in data mining, machine learning, DBMS, and data structure. Dr. Mittal is a
lifetime member of CSI and has published more than 80 research papers in SCI, SCIE,
and Scopus-indexed journals. She holds five patents, two copyrights granted, and three
more published patents in the areas of artificial intelligence, IoT, and deep learning. Dr.
Mittal has edited/authored many books with reputed publishers like Springer, IOS Press,
Elsevier, and CRC Press and is working on a DST-approved project “Development of
IoT-Based Hybrid Navigation Module for Mid-sized Autonomous Vehicles” with a research
grant of 25 lakhs. Currently, she is guiding PhD scholars in the areas of machine learning,
computer vision, and deep learning. Dr. Mittal is an editorial board member for publishers
like Inderscience, Bentham Science, Springer, and Elsevier, and has handled special issues
and chaired a number of conferences. She is book series editor of Innovations in Health
Informatics and Healthcare: Using Artificial Intelligence and Smart Computing and
another series Edge AI in Future Computing for CRC Press, Taylor & Francis, United States.
She is associate editor, advisory member, and editor for Springer journals, Dyna (Spain),
and Elsevier journals, respectively.
This page intentionally left blank
Preface

Machine learning (ML) techniques are used as predictive models for many applications
including those in the field of biomedicine. These techniques have shown impressive
results across a variety of domains in biomedical engineering research. Biology and med-
icine are data-rich disciplines, but the data are complex and often not properly under-
stood. Most biomedical data are categorized into structured, semi-structured, and
unstructured types with very high volume. The volume and complexity of these data pre-
sent new opportunities, but also pose new challenges. Automated algorithms that extract
meaningful patterns could lead to actionable knowledge and change how we develop
treatments, categorize patients, or study diseases, all within privacy-critical environ-
ments. This book addresses the issues described to predict and model biomedical data
mining and analysis. The book has been organized into 15 chapters.
Chapter 1 titled “Data Mining with Deep Learning in Biomedical Data” presents a time-
domain study of EEG signals to detect various neurological disorders with a specific focus
on epilepsy. The presented convolutional neural network (CNN), long short-term memory
network (LSTM), and CNN-LSTM hybrid models were used to detect seizure activities in
precisely filtered EEG segments. The experimental results reveal the suitability of the
CNN-LSTM hybrid model for accurate and prompt detection of epileptic seizures with
an accuracy of 98%, sensitivity of 98.48%, and specificity of 99.19%, so that patients could
be saved from major injuries or sudden expected deaths. These models can be useful in
the detection of various diseases or disorders such as schizophrenia, Parkinson’s disease,
and the identification of breast cancer and bone- or skin-related diseases.
Chapter 2 titled “Applications of Supervised Machine Learning Techniques with the
Goal of Medical Analysis and Prediction: A Case Study of Breast Cancer” analyzes the Wis-
consin Breast Cancer Diagnosis dataset for identifying essential features and assessing the
performance of some popular machine learning (ML) classifiers in breast cancer predic-
tion. The dataset is first cleaned by eliminating non-numerical values and normalizing the
data. The processed data are then visualized to grasp the hidden patterns and non-
essential attributes are trimmed. Eight different ML models are trained and tested over
the refined data for prediction of the two tumor classes. The presented study identified
vital features that were must-haves for the analysis, and the empirical results investigated
the superiority of particular ML classifiers over the others.
Chapter 3 titled “Medical Decision Support System Using Data Mining” describes how
a medical decision support system can support the medical decision-making processes at
both clinical and diagnostic levels. To provide an error-free and accurate service, clinicians

xix
xx Preface

must apply relevant computer-based information and decision support systems. Decision
support systems can be designed as a system based on knowledge or a system based on
learning. Human-engineered mappings to suggestions based on best medical treatments
and patient data are known as knowledge-based systems. Learning-based systems utilize
data mining, statistics, and ML approaches to map the system. Integrated decision sup-
port incorporates both the systems of knowledge and learning to solve the problem of
presence of partial information in a realistic situation. This effort aims to assist physicians
medically and to apply the medicine prescription specifically. The approach may be
utilized for query-based applications, online web browser applications, or mobile appli-
cations on numerous terminal interfaces.
Chapter 4 titled “Role of AI Techniques in Enhancing Multi-Modality Medical Image
Fusion Results” outlines the benefits of using AI methods for medical image fusion of dif-
ferent modalities. The modality can be computed tomography, magnetic resonance-T1,
magnetic resonance-T2, and Positron emission tomography depending on the suspected
malignant region. The aim of fusion is to collaborate each modality’s best information into
a single image called a fused image. This chapter addresses the multi-modality medical
image fusion using AI techniques like Fuzzy Logic and Adaptive Neuro-Fuzzy Inference
System (ANFIS). The study reveals that the AI techniques not only give better results
but their learning capabilities will likely make the future work self-driven.
Chapter 5 titled “A Comparative Performance Analysis of Backpropagation Training
Optimizers to Estimate Clinical Gait Mechanics” indicates that the clinical gait analysis
of healthy people of different age groups plays a significant role in the early estimation
of different physiological and neurological disorders. However, due to complicated data
acquisition setups and in-person requirements, the estimation of the gait analysis has
been quite tough to follow. To avoid such issues, a ML-based approach has been proposed
in this work to estimate the biomechanical gait parameters. Three backpropagation neural
network models with Levenberg-Marquardt method, resilient backpropagation method,
and gradient descent method optimizers have been designed to estimate the joint angles,
joint moments, and ground reaction forces in the sagittal plane. The dataset used in the
neural network models has been taken from an open-source repository. The anthropo-
metric, biological, and spatiotemporal parameters of 50 different subjects have been
exploited as input dataset.
Chapter 6 titled “High-Performance Medicine in Cognitive Impairment: Brain–Computer
Interfacing for Prodromal Alzheimer’s Disease” suggests that Alzheimer’s disease is fre-
quently misdiagnosed as normal aging because it has always been difficult to detect early
on. Mild cognitive impairment (MCI) can be identified, but there is little that can be
done at that time because no medicine can reverse the effect of MCI; instead, it can only
slow down the progression. Alzheimer’s disease is difficult to diagnose medically, espe-
cially in its early stages. As a response, a method for early diagnosis of Alzheimer’s dis-
ease is urgently needed even now. In this chapter, the authors have proposed a strategy
for detecting Alzheimer’s disease in its early stage using noninvasive brain-computer
interface technology. Electroencephalography (EEG) brain wave patterns were used for
 Preface xxi

three groups (Alzheimer’s disease—AD, mild cognitive impairment—MCI, and healthy

subjects—HS) of test subjects in this research. The proposed framework was evaluated
with 46 test subjects, with an accuracy of 86.47% and a precision of 0.801.
Chapter 7 titled “Brain Tumor Classifications by Gradient and XG Bosting Machine
Learning Models” describes the use of the boosting-type ML algorithms to evaluate the
model performance parameters. Model performance is validated using K-fold methods
and preliminary results indicate that the XG boosting algorithm yields the highest classi-
fication accuracy. Evaluations of this type are largely supportive of biomedical imaging
studies and there is scope for future studies using other classification models for achieving
the highest prediction accuracy.
Chapter 8 titled “Biofeedback Method for Human–Computer Interaction to Improve
Elder Caring: Eye-Gaze Tracking” proposes how physiological methods of eye-gaze track-
ing could be used to design and develop natural user interaction techniques. A human
user’s tacit intention to use physiological signals for the domestic area’s required activi-
ties/requirements may be understood by utilizing nonverbal contact to define the user’s
intention to use physiological signals for the domestic area’s necessary activities/require-
ments. To achieve good accuracy and robustness, traditional gaze monitoring systems
depend on explicit infrared lights and high-resolution cameras. Recent advancements
in mobile devices, as well as an increasing interest in recording normal human behavior,
have shown that tracking eye motions in a non-restricted environment could yield prom-
ising results.
Chapter 9 titled “Prediction of Blood Screening Parameters for Preliminary Analysis
Using Neural Networks” describes various techniques used in the prediction of blood
parameters. The prediction of blood screening test features using the backpropagation
neural network is presented in detail. The features used in this chapter were fibrinogen
and globulin. The normal ranges of fibrinogen and globulin are 2–4 g/L and 20–35 g/L,
respectively. Fibrinogen is a glycoprotein that circulates in the blood of all vertebrates.
It is observed from the results that the prediction accuracy for fibrinogen is better than
that for globulin. To increase the accuracy of the prediction for globulin, the training
parameters and activation functions must be modified.
Chapter 10 titled “Classification of Hypertension Using an Improved Unsupervised
Learning Technique and Image Processing” presents an improved nearest neighbor dis-
tance clustering algorithm by recognizing the lesions present in the retina. The current
approach identifies the symptoms associated with retinopathy for hypertension and clas-
sifies the hypertensive retinopathy. This chapter provides an assessment of the hyperten-
sive retinopathy recognition techniques that apply a range of image processing
procedures used for feature extraction and classification. The chapter also specifies the
existing open databases, containing eye fundus images, which can be used for hyperten-
sive retinopathy research.
Chapter 11 titled “Biomedical Data Visualization and Clinical Decision-Making in
Rodents Using a Multi-usage Wireless Brain Stimulator With a Novel Embedded Design”
describes in detail the complete design, biomedical data visualization, and modeling
xxii Preface

aspects of the stimulator device. The feasibility of this device is successfully tested in
in vivo and in vitro stages for a period of more than a month. This embedded design
has been developed taking into account cost-effectiveness, user-friendliness, and preci-
sion, which are the main focus of this chapter. The brain-computer interface can be useful
in taking effective clinical decision-making at an early stage. However, there is limited
research in this area so far. Therefore, all the efforts in this direction are extremely impor-
tant for numerous young flourishing specialists, and aspirations toward the brain-
computer interface.
Chapter 12 titled “LSTM Neural Network-Based Classification of Sensory Signals for
Healthy and Unhealthy Gait Assessment” describes the modeling of the long short-term
memory (LSTM) deep neural network model and its implementation to classify healthy
and unhealthy gait based on a sensory dataset. The reference sensory dataset of 22 subject
samples (11 healthy and 11 with knee pathology) is taken from the UCI Irvine Machine
Learning Repository. Two different optimizers, namely Stochastic Gradient Descent and
Adam, have been exploited in the designed LSTM model with different sets of learning
hyperparameters. The classification results of the proposed deep learning model with
both optimizers have been compared with each other using several performance mea-
sures like precision, recall, and F1 score.
Chapter 13 titled “Data-Driven Machine Learning: A New Approach to Process and
Utilize Biomedical Data” includes a study of precise and accurate diagnostic tools to ease
the pressure on medical personnel, simultaneously enhancing efficiency. This chapter
explores the development of artificial neural network based diagnostic tools that focus
on the challenges described previously. A brief overview of the current scenarios and
future prospects of ML in biomedicine is also presented.
Chapter 14 titled “Multiobjective Evolutionary Algorithm Based on Decomposition for
Feature Selection in Medical Diagnosis” presents a mathematical model of a multi-
objective evolutionary algorithm based on decomposition (MOEA/D) and its application
in feature selection in medical diagnosis. Most of the medical datasets are high dimen-
sional in nature and so there is a need for optimal feature selection, which is a difficult
problem. The negative influence may be due to the possibility of irrelevant or many redun-
dant features. Intelligent models including classification, clustering, regression, and
boosting techniques are helpful in extracting useful knowledge. The performance of
the MOEA/D method is compared with that of state-of-the-art multi-objective optimiza-
tion methods when applied to most of the datasets.
Chapter 15 titled “Machine Learning Techniques in Healthcare Informatics: Showcas-
ing Prediction of Type 2 Diabetes Mellitus Disease using Lifestyle Data” focuses on the role
of the ML paradigms in healthcare analytics and presents the implementation of the
framework for developing ML models for type 2 diabetes mellitus (T2DM) disease. In this
chapter, lifestyle indicators rather than clinical/pathological parameters have been used
for the prediction of T2DM. The study involves different experts like diabetologists, endo-
crinologists, dieticians, and nutritionists for selecting the contributing lifestyle parame-
ters to promote health and manage diabetes. The study aims to develop an intelligent
 Preface xxiii

knowledge-based system for the prediction of T2DM without conducting clinical tests. It
can save the patient undue delays caused by unnecessary readmissions and pathological
tests in hospitals. The proposed work emphasizes the use of ML techniques, namely K-
nearest neighbor (KNN), logistic regression (LR), naı̈ve Bayes (NB), support vector
machine (SVM), decision tree (DT), random forest (RF), and artificial neural network
(ANN), for the prediction of T2DM disease. The RF technique attained the highest accu-
racy of 93.56% followed by DT, LR, SVM, NB, ANN, and KNN with accuracies of 92.70%,
91.41%, 90.98%, 89.27%, 87.98%, and 84.54%, respectively.
We are grateful to Elsevier, especially Chris Katsaropoulos, Senior Acquisitions Editor,
for providing us the opportunity to edit this book.

Sudipta Roy
Maharashtra, India
Lalit Mohan Goyal
Faridabad, India
Valentina E. Balas
Romania
Basant Agarwal
Jaipur, India
Mamta Mittal
New Delhi, India
This page intentionally left blank
 1
Data mining with deep learning
in biomedical data
Kuldeep Singha and Jyoteesh Malhotrab
a
DEPARTMENT OF ELECTRONICS TECHNOLOGY, G U R U NAN A K DE V UN I V E R S I T Y AM R I T S AR ,
P U NJAB, I NDI A b DEPARTMENT OF ENGI NEERING AND TECHNO LOGY , GURU NANAK DEV
U N I V E R S I T Y RE GI ON A L C AM P U S JALANDHAR, PUNJAB, INDIA

1. Introduction
In the era of Internet of things (IoT) technologies, smart health care is an emerging sector
that is attracting the attention of medical personnel, the research community, and
patients [1, 2]. These technological advances in association with machine learning and
cloud-fog computing capabilities have started revolutionizing the health-care sector by
shifting the traditional patient monitoring approach to remote patient monitoring [3].
In this sector, biomedical data analysis is crucial in the detection and diagnosis of a variety
of health-related issues such as bacterial and viral infectious diseases; neurological and
mental disorders, particularly, epilepsy, schizophrenia, Alzheimer’s disease, etc.; cardio-
vascular diseases; autoimmune diseases; cancer; and skin- or bone-related diseases [2,
4–10]. The biomedical data may include electroencephalogram (EEG) or electrocardio-
gram (ECG) signals, X-ray, CT scan, MR-based images or microscopic images, etc., which
could be analyzed using machine learning or deep learning-based signal analysis tech-
niques [10–13].
Among the aforementioned diseases, neurological and mental disorders are one of the
most serious hazards to public health [4]. These disorders have become one of the main
causes of disabilities and deaths globally. The social and economic burden of these disor-
ders is more severe in underdeveloped or impoverished countries due to a scarcity of
health-care infrastructure. This burden is likely to grow rapidly in forthcoming years as
a result of an intensive increase in population and aging [14]. Modern IoT-enabled
health-care technologies may be useful in detecting and predicting these neurological
and mental problems to save patients’ lives.
In this regard, epilepsy is considered to be a prevalent fatal neurological illness, which
usually affects the human brain by causing spontaneous and repeated seizures [15]. This
disorder distresses the lives of different age groups from infants to old people. According
to the World Health Organization (WHO) data, around 50 million individuals suffer from
epilepsy around the world [16], and 80% of them are living in the developing and under-
developed countries. According to the Indian Epilepsy Centre in New Delhi, around

Predictive Modeling in Biomedical Data Mining and Analysis. https://doi.org/10.1016/B978-0-323-99864-2.00018-4 1

Copyright © 2022 Elsevier Inc. All rights reserved.
2 Predictive Modeling in Biomedical Data Mining and Analysis

10 million Indians are the victims of this illness and this number is growing by the day with
the annual addition of 0.5–1 million new epileptic patients [17]. All these facts make it one
of the most common global neurological diseases and a major threat to mankind.
The usual rhythm of neuronal functions in the human brain is disrupted during
epileptic seizures, resulting in crucial clinical symptoms such as erratic behavior, weird
stimuli, emotions, cramps, and reduced or lost consciousness, among others [15, 18].
These factors may result in serious injuries or unexpected deaths of patients as a result
of brain injury or suffocation in traffic accidents or while working in dangerous work envi-
ronments [15]. Epileptic seizures can be managed and controlled by practicing regular
meditation activities and taking antiepileptic drugs [19, 20]. In some cases, when these
therapies fail to control the onset of seizures and the patients become drug resistant, then
the brain surgeries could be performed to remove seizure-causing neurons from their
brain for reducing the occurrence of seizures [21]. But all these remedies are not able
to provide any everlasting cure to this disorder. Therefore, there is a pressing need for
an automated system of epileptic seizure detection that can detect seizure onsets, and
can transfer alert messages to the family, doctors/neurologists, or ambulance services
to save the lives of epileptic patients [22].
The efficient detection of epileptic seizures could be achieved by brain monitoring of
the patients using EEG technique [23–26]. It is because of the capability of EEG to monitor
brain functions directly from the brain and generate signals with high temporal and
spatial resolutions [10, 27]. Furthermore, the use of machine learning and deep learning
techniques has been currently being investigated by the research community for timely
identification of seizure initiation using EEG signal processing [22, 28]. Deep learning
techniques, instead, have grown in popularity for accurate seizure detection due to their
independence from feature extraction techniques and ability to efficiently handle issues
such as big data and the nonstationary existence of EEG signals [7, 11, 29].
In this direction, this chapter also focuses on exploring the utilization of deep learning
techniques for data mining on biomedical signals like EEG, to find effective solutions in
the health-care domain. This chapter primarily focuses on the use of deep learning tech-
niques, namely convolutional neural network (CNN), long short-term memory network
(LSTM), and CNN + LSTM hybrid model to detect seizure activities through EEG signal
analysis in the time domain. In the proposed approach of seizure detection, EEG signals
are undergone through filtering to avoid the effect of noise and segmentation to make
them stationary. Afterwards, these segments are fed to the proposed architectures of deep
learning algorithms for classification. The analysis of simulation results in terms of various
performance measures ensures the suitability of CNN-LSTM model to accurately detect
seizure activities of the patients with epilepsy in real time.
This chapter has been organized into various sections. Section 1 debates on the
emergence of the health-care sector and various disorders along with their detection
and management with a special focus on epilepsy. The role of deep learning approaches
for the detection of epileptic seizures has been elaborated in Section 2. Furthermore,
Section 3 presents the proposed method of seizure detection using deep learning models
 Chapter 1 • Data mining with deep learning in biomedical data 3

using EEG signal analysis in time domain. The simulation results of this approach are
demonstrated and debated in Section 4. Finally, Section 5 delivers the conclusions made
after analyzing the outcomes of the proposed approach.

2. Role of deep learning techniques in epileptic seizure

detection
Deep learning methods have excelled in a wide range of applications, including computer
vision and natural language processing [30–32]. Nowadays, these methods have begun to
play a significant role in health-care applications [33]. To address the limitations of
conventional machine learning algorithms, the research community is turning to deep learn-
ing techniques for epileptic seizure detection [34]. Traditional machine learning methods are
unsuitable for real-time deployment due to their inability to manage large amounts of EEG
data and their long processing times [35]. Deep learning techniques are also not dependent
upon feature extraction methods, unlike conventional machine learning techniques, and are
well suited to deal with the nonstationary character of EEG signals [36–38].
However, due to the black-box nature of deep learning algorithms caused by overpar-
ameterization, their prediction results are difficult to grasp, thereby resulting in lack of
interpretability [39, 40]. But there exists a variety of interpretation algorithms, which could
assess the trustworthiness of given deep learning algorithms. Thus, the comparison and
evaluation of outcomes of interpretation algorithms with the output estimates of human
experts lead to the interpretability of deep learning models [39, 41, 42]. Furthermore,
another issue in deep learning algorithms is the availability of small datasets in the
health-care field [43, 44]. In the era of big data, the problem of small data could be man-
aged by training state-of-the-art classifier models with given data in an unsupervised or
semisupervised manner, for example, using generative adversarial networks, autoenco-
ders, etc. [43], as well as using the concept of transfer learning [44, 45]. Many researchers
have obtained optimal classification performance from deep learning algorithms using
small datasets in various health-care applications [46–50].
This section explores various seizure detection systems that use deep learning tech-
niques to analyze EEG signals in order to examine their capabilities. Since the present
work makes use of Bonn dataset of EEG signals [51], developed at the University of Bonn,
Germany, most of the literature mentioned in the following paragraphs is based on deep
learning-based epileptic seizure detection using the same dataset.
Acharya et al. [52] presented a 13-layer deep CNN classifier for automatic detection of
seizures using EEG data, which had accuracy value, sensitivity value, and specificity value
of 88.67%, 95%, and 90%, respectively, using the influence of deep learning. In the same
concern, Singh and Malhotra [53] had proposed a stacked autoencoder model, which took
into account short-duration EEG segments for seizure detection and provided results with
accuracy value of 88.8%, the sensitivity value of 89.44%, and specificity value of 93.77%. An
LSTM was also applied in another technique for effective epileptic seizure detection [54],
which provided 95.54% accuracy. A technique presented by Singh and Dehuri [55]
4 Predictive Modeling in Biomedical Data Mining and Analysis

decomposed EEG signals using discrete wavelet transform (DWT) and fed them to mul-
tilayer perceptron (MLP)-based deep learning algorithm, which provided accuracy value
of 100% for binary classification of seizure and nonseizure classes.
Furthermore, a model based on a stacked bidirectional LSTM network (Bi-LSTM) [56]
performed well at detecting seizures, with the accuracy value of 99.08%, precision value
of 98%, and recall value of 99.5%. Another LSTM-based method for detecting seizure activity
used short-duration temporal segments of EEG signals [57]. To keep this approach compu-
tationally efficient, these EEG segments only contained two data points. It conducted aver-
age pooling of hidden layer outputs from various LSTM devices, which were then fed to
SoftMax activation-based classification layer for classification. For three-class classification,
this model achieved 100% classification accuracy, sensitivity, and specificity. A new CNN
approach presented by Tu €
€rk and Ozerdem [58] made use of continuous wavelet transform
(CWT) to obtain two-dimensional (2D) frequency-time scalograms of EEG signals and
attained an accuracy of 99% for three-class classification. Similarly, a CNN-based model
for classification of seizure states used segmented EEG signals with a fixed size overlapping
window and had an average accuracy of 99.1% for detecting seizure behaviors [59].
Akut [60] suggested a DWT-based technique for detecting epileptic seizures using CNN
in the same context. This method used DWT to decompose EEG signals into five levels,
which were provided to the proposed CNN model for classification. For a three-class prob-
lem, it produced classification with accuracy value of 99.4%, sensitivity value of 98.5%, and
specificity value of 98.45%. Moreover, for seizure state classification, another CNN model
used raw EEG signals and presented classification results with a 99% accuracy [61]. Daoud
and Bayoumi [62] also presented a deep convolutional autoencoder with MLP combined
model for seizure classification using EEG signals, which had a 96% accuracy, 93% sensi-
tivity, and 99% specificity.
Similarly, several recently published papers describe methods for detecting seizures
effectively. Ilakiyaselvan et al. [63] proposed a seizure detection technique that used
the reconstructed phase space (RPS) approach to transform EEG signals into images. Alex-
Net, a pretrained CNN model, was considered to classify the RPS images into different sei-
zure states. For three-class classification, this technique had a 95% accuracy rate. Another
deep learning technique for three-class seizure classification made use of CNN-LSTM
series model with EEG segments and got a classification accuracy of 95.9%. In this context,
Glory et al. [64] had discussed the idea of adaptive Haar wavelet-based binary grasshopper
optimization algorithm for processing of EEG signals and classification using deep neural
network (DNN). They obtained accuracy value of 93.61%, sensitivity value of 90.24%, and
specificity value of 93.63% for seizure detection in the case of Bonn dataset. Furthermore,
Zhao et al. [65] explored another CNN-based method for seizure classification that used
raw time-domain EEG segments. In a three-class problem, this approach achieved 96.97%
classification accuracy for the detection of seizures.
Moreover, Woodbright et al. [66] had proposed a novel technique for seizure detection
using CNN-based feature extraction, which provided classification accuracy value of
98.65%, sensitivity value of 96.29%, and specificity value of 99.25%. Similarly,
 Chapter 1 • Data mining with deep learning in biomedical data 5

Ramakrishnan et al. [67] had presented one-dimensional adaptive average local binary
pattern (1D AaLBP) and one-dimensional adaptive difference local binary pattern (1D
AdLBP) for extracting features and CNN for classification. This method obtained maxi-
mum classification accuracy of 98.88% for three-class seizure classification.
The literature review of deep learning techniques for seizure detection is summarized
in Table 1. These techniques provide higher accuracies for the desired task and are well
suited for real-time use due to their ability to manage the nonstationary character of EEG.

3. Proposed method of seizure detection

This section debates the proposed method of detecting seizure activities in the patients with
epilepsy and the methodology adopted for its execution. Various steps employed in this
model are presented in Fig. 1, which includes EEG dataset collection, filtering, segmenta-
tion, and deep learning-based classification. This model has been implemented using
Python language on a personal computer having a configuration of Intel i7 processor with
16 GB RAM and Nvidia GEFORCE GTX 1060 6 GB GPU with Windows 10 operating system.
The publicly accessible EEG dataset of the University of Bonn, Germany, was used in
this study [51]. It is made up of EEG segments belonging to five subsets, denoted by the
letters A, B, C, D, and E. These subsets contain 100 single-channel EEG samples, where
each sample has a segment duration of 23.6 s. The standard electrode placement tech-
nique [69] is used to measure each EEG sample, which consists of 14 surface electrodes
positioned on the patient’s skull. These EEG data samples were captured by means of a
128-channel amplifier and were digitized using a 12-bit analog to digital converter with
173.61 Hz sampling rate. The given EEG samples are classified as normal, preictal, and
ictal seizure classes. The subsets A and B are acquired from the scalps of five healthy indi-
viduals and are represented as “normal” among the given five subsets. The remaining sub-
sets C, D, and E were collected from the scalps of five patients suffering with epilepsy. The
subsets C and D, which are described as “interictal,” are made up of EEG samples obtained
from epileptic and opposite hemispheres of the brain during nonseizure periods. EEG
samples only linked to seizure activities are represented as “ictal” in subset E. As a result,
the given dataset of 500 samples is divided into three groups: A–B (normal) with 200 sam-
ples, C–D (interictal) with 200 samples, and E (ictal) with 100 samples. Fig. 2 shows the
plots of EEG signals at normal class for healthy people and interictal and ictal classes
for epileptic people.
During the acquisition process, due to the movement of the patients’ eyes or body
parts, and during transmission, due to the superimposition of channel noise or defective
connecting devices, the EEG signals are disrupted by different types of artifacts or noises
[70, 71]. These tainted EEG signals may lead to incorrect seizure detection and should be
properly filtered. To keep raw EEG signals free of noise-causing undesirable frequency
components, they are filtered by means of a second-order Butterworth filter, which is
designed to operate on a lower cutoff frequency of 0.1 Hz and a higher cutoff frequency
of 70 Hz in the present work.
Another Random Scribd Document
with Unrelated Content
Venegas to join them, they would have got little profit out of having
23,000 more Spaniards on the Alberche. Sebastiani on the other
hand, who could join Victor at the same moment that the corps from
La Mancha joined the allies, would bring some 17,000 excellent
troops to Talavera. The benefit of drawing in Venegas would be
much less than the disadvantage of drawing in Sebastiani to the
main theatre of war. Hence came the idea that the army from the
Passes must be devoted to the sole purpose of keeping the 4th
Corps as far as possible from the Alberche. Even knowing that
Venegas was hostile to Cuesta, and that he was a man of no mark or
capacity, Wellesley could not have expected that he would disobey
orders, waste time, and fail utterly in keeping touch with Sebastiani
or threatening Madrid.
The one irreparable fault in the drawing up of the whole plan of
campaign was the fundamental one that Wellesley had undertaken
to co-operate with Spanish armies before he had gauged the weak
points of the generals and their men. If he had held the post of
commander-in-chief of the allied forces, and could have issued
orders that were obeyed without discussion, the case would have
been different. But he had to act in conjunction with two colleagues,
one of whom was suspicious of his intentions and jealous of his
preponderant capacity, while the other deliberately neglected to
carry out clear and cogent orders from his superior officer. Cuesta’s
impracticability and Venegas’s disobedience could not have been
foreseen by one who had no previous experience of Spanish armies.
Still less had Wellesley realized all the defects of the Spanish rank
and file when placed in line of battle. That he did not hold an
exaggerated opinion of their merits when he started on the
campaign is shown by letters which he wrote nine months
before[603]. But he was still under the impression that, if cautiously
handled, and not exposed to unnecessary dangers, they would do
good service. He had yet to witness the gratuitous panic of Portago’s
division on the eve of Talavera, and the helplessness of the Spanish
cavalry at the combats of Gamonal and Arzobispo. After a month’s
experience of Cuesta and his men, Wellesley vowed never again to
take part in grand operations with a Spanish general as his equal
and colleague. This was the teaching of experience—and on July 10
the experience was yet to come.
The interview at the bridge of Almaraz had not been very
satisfactory to Wellesley, but it was far from having undeceived him
as to the full extent of the difficulties that lay before him. He wrote
to Frere at Seville that he had been on the whole well received, and
that Cuesta had not displayed any jealousy of him. As that sentiment
was at this moment the predominant feeling in the old man’s breast,
it is clear that he had succeeded in hiding it. But the obstinate
silence of Wellesley’s colleague had worried him. O’Donoju had done
all the talking, and ‘it was impossible to say what plans the general
entertains.’ He was moreover somewhat perturbed by the rumours
which his staff had picked up from the Estremaduran officers, to the
effect that Cuesta was so much the enemy of the Central Junta that
he was plotting a pronunciamiento for its deposition[604]. As to the
fighting powers of the Spanish army, Wellesley wrote to Castlereagh
that ‘the troops were ill clothed but well armed, and the officers
appeared to take pains with their discipline. Some of the corps of
infantry were certainly good, and the horses of the cavalry were in
good condition.’ Only ten days later he was to utter the very different
opinion that ‘owing to their miserable state of discipline and their
want of officers properly qualified, these troops are entirely
incapable of performing any manœuvre however simple[605],’ and
that ‘whole corps, officers and men, run off on the first appearance
of danger[606].’
The British Commander-in-chief had indeed many moral and
mental experiences to go through between the interview at Mirabete
on July 10, and the retreat from Talavera on August 2!
 SECTION XVI: CHAPTER IV
THE MARCH TO TALAVERA: QUARREL OF
WELLESLEY AND CUESTA

Having returned to his army on July 12, Wellesley gave orders for
the whole force to get ready for a general advance on the morning
of the eighteenth, the day which had been chosen for the
commencement of operations at the conference of Almaraz. It would
have been in every way desirable to have moved out at once, and
not to have waited for these six days. If the march against Victor
had been fixed for the thirteenth or fourteenth, the French would
have been caught unprepared, for as late as the seventeenth King
Joseph and his adviser Jourdan were under the impression that the
force at Plasencia consisted of nothing more than a Portuguese
division of 10,000 men, and it was only on the twenty-second that
they received the definite information that the whole British army
was upon the Tietar[607]. It is clear that, by advancing five days
earlier than he actually did, Wellesley might have caught the enemy
in a state of complete dispersion—the 4th Corps being on July 20
still at Madridejos in La Mancha, and the King with his reserves at
Madrid. If attacked on the seventeenth or the eighteenth, as he
might well have been, Victor would have found it impossible to call
up Sebastiani in time, and must have fallen back in haste to the
capital. The allies could then have cut him off from the 4th Corps,
which must have retreated by a circuitous route, and could not have
rejoined the main body of the French army in time for a battle in
front of Madrid.
It would appear that Wellesley had fixed the date of his advance
so late as the eighteenth mainly because of the difficulty as to the
collection of provisions, which was now looming before him in larger
proportions than ever. But it is possible that the necessity for
allowing some days for the transmission of the plan of campaign to
Venegas also counted for something in the drawing up of the time-
table. It would have been rash to start before the army of La
Mancha was prepared to take its part in the joint plan of operations.
So much depended upon the diversion which Venegas was to
execute, that it would have been a mistake to move before he could
break up from his distant cantonments at Santa Cruz de Mudela. No
word, however, concerning this appears in Wellesley’s
correspondence. From July 13 to July 18 his dispatches show anxiety
about nothing save his food and his transport. Every day that he
stayed at Plasencia made him feel more uncomfortable concerning
the all-important question of supplies. The corn which the Alcaldes
of the Vera had promised to secure for him had begun to come in,
though in driblets and small consignments, but there was no means
of getting it forward: transport was absolutely unprocurable[608].
Wellesley sent officers to scour the country-side as far as Bejar and
Ciudad Rodrigo, but they could procure him neither mules nor carts.
He also pressed the Spanish commissary-general, Lozano de Torres,
to hunt up every animal that could be procured, but to small effect.
The fact was that Estremadura was not at any time rich in beasts or
vehicles, and that the peasantry had sent away most of those they
owned while the French lay at Almaraz, lest they should be carried
off by the enemy. Wellesley, who did not understand the limited
resources of this part of Spain, was inclined to believe that the
authorities were hostile or even treacherous. The Central Junta had
promised him transport in order to make sure of his starting on the
campaign along the Tagus, and when transport failed to appear, he
attributed it to ill-will rather than to poverty. No doubt he was fully
justified in his view that an army operating in a friendly country may
rationally expect to draw both food and the means to carry it from
the regions through which it is passing. But sometimes the
provisions or the transport are not forthcoming merely because the
one or the other is not to be found. It is certain that both
Estremadura and the valley of the central Tagus were at this
moment harried absolutely bare: Victor’s despairing letters from
Caceres in May and from La Calzada in June are sufficient proof of
the fact. In a district where the Marshal said that ‘he could not
collect five days’ provisions by any manner of exertion,’ and that ‘his
men were dropping down dead from actual starvation, so that he
must retire or see his whole corps crumble away[609],’ it is clear that
the Central Junta could not have created food for the British army.
Cuesta’s troops were living from hand to mouth on supplies sent
forward from Andalusia, or they could not have continued to exist in
the land. The only district which was intact was that between Coria
and Plasencia, and this was actually at the moment feeding the
British army, and had done so now for ten days or more. But
unfortunately the Vera could give corn but no draught animals. If
Wellesley had known this, he must either have exerted himself to
procure more transport before leaving Abrantes—a difficult task, for
he had already drained Portugal of carts and mules—or have refused
to march till the Spaniards sent him wagon trains from Andalusia. It
would have taken months for the Junta to collect and send forward
such trains: they had dispatched all that they could procure to
Cuesta. The campaign on the Tagus, in short, would never have
been fought if Wellesley had understood the state of affairs that he
was to encounter.
The causes, therefore, of the deadlock that was about to occur
were partly the light-hearted incompetence of the Central Junta in
promising the British army the use of resources which did not exist,
partly Wellesley’s natural ignorance of the miserable state of Central
Spain. He had never entered the country before, and could not know
of its poverty. He had trusted to the usual military theory that the
country-side ought to provide for a friendly army on the march: but
in Spain all military theories failed to act. Napoleon committed
precisely similar errors, when he directed his army corps to move
about in Castile as if they were in Germany or Lombardy, and found
exactly the same hindrances as did the British general. In later years
Wellesley never moved without a heavy train, and a vast provision of
sumpter-beasts and camp-followers. In July 1809 he had still to
learn the art of conducting a Spanish campaign.
Meanwhile he was beginning to feel most uncomfortable about
the question of provisions. His anxiety is shown by his letters to
Frere and Beresford; ‘it is impossible,’ he wrote, ‘to express the
inconvenience and risk that we incur from the want of means of
conveyance, which I cannot believe the country could not furnish, if
there existed any inclination to furnish them. The officers complain,
and I believe not without reason, that the country gives unwillingly
the supplies of provisions that we have required ... and we have not
procured a cart or a mule for the service of the army[610].’ But to
O’Donoju, the chief of the staff of the Estremaduran army, he wrote
in even more drastic terms, employing phrases that were certain to
provoke resentment. He had, he said, scoured the whole region as
far as Ciudad Rodrigo for transport, and to no effect. ‘If the people
of Spain are unable or unwilling to supply what the army requires, I
am afraid that they must do without its services.’ He had been forced
to come to a painful decision, and ‘in order to be fair and candid to
General Cuesta’ he must proceed to inform him that he would
execute the plan for falling upon Victor behind the Alberche, but that
when this had been done he would stir no step further, and ‘begin
no new operation till he had been supplied with the means of
transport which the army requires[611].’
After dispatching this ultimatum, whose terms and tone leave
something to be desired—for surely Cuesta was the last person to be
saddled with the responsibility for the pledges made by his enemies
of the Central Junta—Wellesley issued orders for the army to march.
He had been joined at Plasencia by the last of the regiments from
Lisbon, which reached him in time for Talavera[612], but had been
forced to leave 400 sick behind him, for the army was still in a bad
condition as regards health. It was therefore with little over 21,000
men that he began his advance to the Alberche. It was executed
with punctual observance of the dates that had been settled at the
interview at Almaraz. On July 18 the army crossed the Tietar on a
flying bridge built at Bazagona, and lay at Miajadas. On the next
night the head quarters were at Centinello; on the twentieth the
British entered Oropesa. Here Cuesta joined them with his whole
army, save the two battalions lent to Wilson, and the two others
under the Marquis Del Reino which had been sent to the Puerto de
Baños. Deducting these 2,600 bayonets and his sick, he brought
over 6,000 horse and 27,000 foot to the rendezvous. The junction
having taken place on the twenty-first, the advance to Talavera was
to begin next morning. Oropesa lies only nineteen miles from that
town, and as Victor’s cavalry vedettes were in sight, it was clear that
contact with the enemy would be established during the course of
the day. Accordingly the allied armies marched with caution, the
Spaniards along the high-road, the British following a parallel path
on the left, across the slopes of the hills which divide the valley of
the Tietar from that of the Tagus.
About midday the Spaniards fell in with the whole of the cavalry
division of Latour-Maubourg, which Victor had thrown out as a
screen in front of Talavera. He had ascertained on the evening of the
preceding day that Cuesta was about to move forward, and was
anxious to compel him to display his entire force. Above all he
desired to ascertain whether the rumours concerning the presence
of British troops in his front were correct. Accordingly he had left
two battalions of infantry in the town of Talavera, and thrown out
the six regiments of dragoons in front of it, near the village of
Gamonal. The Spaniards were advancing with Albuquerque’s cavalry
division as an advanced guard. But seeing Latour-Maubourg in his
front the Duke refused to attack, and sent back for infantry and
guns. Cuesta pushed forward the division of Zayas to support him,
but even when it arrived the Spaniards made no headway. They
continued skirmishing for four hours[613] till the British light cavalry
began to appear on their left. ‘Though much more numerous than
the enemy,’ wrote an eye-witness, ‘they made no attempt to drive
him in, but contented themselves with deploying into several long
lines, making a very formidable appearance. We had expected to see
them closely and successfully engaged, having heard that they were
peculiarly adapted for petty warfare, but we found them utterly
incapable of coping with the enemy’s tirailleurs, who were driving
them almost into a circle.’
On the appearance, however, of Anson’s cavalry upon their flank
the French went hastily to the rear, skirted the suburbs of Talavera,
and rode off along the great Madrid chaussée to the east, followed
by the British light dragoons. As they passed the town two small
columns of infantry came out of it and followed in their rear.
Albuquerque sent one of his regiments against them, but could not
get his men to charge home. On three separate occasions they came
on, but, after receiving the fire of the French, pulled up and fell into
confusion. The impression made by the Spanish cavalry on the
numerous British observers was very bad. ‘No men could have more
carefully avoided coming to close quarters than did the Spaniards
this day[614],’ wrote one eye-witness. ‘They showed a total lack not
only of discipline but of resolution[615],’ observes another.
After crossing the plain to the north of Talavera the French, both
cavalry and infantry, forded the Alberche and halted on the further
bank. On arriving at the line of underwood which masks the river the
pursuers found the whole of Victor’s corps in position. The thickets
on the further side were swarming with tirailleurs, and two batteries
opened on Anson’s brigade as it drew near to the water, and sent
balls whizzing among Wellesley’s staff when he pushed forward to
reconnoitre the position.
It was soon seen that Victor had selected very favourable
fighting-ground: indeed he had been staying at Talavera long
enough to enable him to get a perfect knowledge of the military
features of the neighbourhood. The 1st Corps was drawn up on a
range of heights, about 800 yards behind the Alberche, with its left
resting on the impassable Tagus, and its right on a wooded hill,
behind which the smaller river makes a sharp turn to the east, so as
to cover that flank. The position was formidable, but rather too long
for the 22,000 men who formed the French army. Having learnt from
the people of Talavera that the enemy had received no
reinforcements up to that morning, from Madrid or any other
quarter, Wellesley was anxious to close with them at once. The
afternoon was too far spent for any attempt to force the passage on
the twenty-second, but on the next day (July 23) the British general
hoped to fight. The Alberche was crossed by a wooden bridge which
the enemy had not destroyed, and was fordable in many places:
there seemed to be no reason why the lines behind it might not be
forced by a resolute attack delivered with numbers which were as
two to one to those of the French.
Accordingly Wellesley left the 3rd division and Anson’s light horse
in front of the right wing of Victor’s position, and encamped the rest
of his army some miles to the rear, in the plain between Talavera
and the Alberche. In the same way Albuquerque and Zayas halted
for the night opposite the bridge on the French left, while the main
body of the Spaniards occupied the town in their rear. In the evening
hours Wellesley endeavoured to urge upon Cuesta the necessity for
delivering an attack at dawn: he undertook to force the northern
fords and to turn the enemy’s right, if his colleague would attack the
southern fords and the bridge. The Captain-General ‘received the
suggestion with dry civility,’ and asked for time to think it over. After
a conference with his subordinates, he at last sent word at midnight
that he would accept the proposed plan of operations.
At 3 o’clock therefore on the morning of the twenty-third,
Wellesley brought down Sherbrooke’s and Mackenzie’s divisions to
the ground opposite the fords, and waited for the arrival of the
Spanish columns on his right. They did not appear, and after long
waiting the British general rode to seek his colleague. He found him
opposite the bridge of the Alberche, ‘seated on the cushions taken
out of his carriage, for he had driven to the outposts in a coach
drawn by nine mules, the picture of mental and physical inability.’
The old man murmured that the enemy’s position had not been
sufficiently reconnoitred, that it would take time to get his army
drawn out opposite the points which it was to attack, that he was
not sure of the fords, that the bridge over which his right-hand
column would have to advance looked too weak to bear artillery, and
many other things to the same effect—finally urging that the forcing
of the Alberche must be put off to the next day. As he had not got
his troops into battle order, it was clear that the morning would be
wasted, but Wellesley tried to bargain for an attack in the afternoon.
The Captain-General asked for more time, and would listen to no
arguments in favour of fighting on that day. After a heated
discussion Wellesley had to yield: he could not venture to assail the
French with his own army alone, and without any assistance from
the Spaniards. Accordingly it was agreed that the advance should
not be made till the dawn of the twenty-fourth.
In the afternoon the pickets sent back information that Victor
seemed to be on the move, and that his line was growing thin.
Cuesta was then persuaded to go forward to the outposts; he was
hoisted on to his horse by two grenadiers, while an aide-de-camp
stood on the other side to conduct his right leg over the croup and
place it in the stirrup. Then, hunched up on his saddle, he rode
down to the river, observed that the greater part of the enemy were
still in position, and refused to attack till next morning.
At dawn, therefore, on the twenty-fourth the allied army moved
forward to the Alberche in three columns, and found, as might have
been expected, that the French had disappeared. On seeing the
masses of redcoats opposite his right upon the previous day, Victor
had realized at last that he had before him the whole British army.
He had sent his train to the rear in the afternoon, and drawn off his
entire force after dusk. By dawn he was more than ten miles away,
on the road to Santa Ollala and Madrid. It was useless to pursue him
with any hope of forcing him to a battle. The chance of crushing him
before he should receive any further reinforcements had
disappeared. It is not at all to his credit as a general that he had
held his ground so long; if he had been attacked on the twenty-
third, as Wellesley had desired, he must certainly have suffered a
disaster. He had but 22,000 men; and it is clear that, while the
Spaniards were attacking his left and centre, he could not have set
aside men enough to hold back the assault of the solid mass of
20,000 British troops upon his right. He should have vanished on the
twenty-second, the moment that Latour-Maubourg reported that
Wellesley’s army was in the field. By staying for another day on the
Alberche he risked the direst disaster.
The British general would have been more than human if he had
not manifested his anger and disgust at the way in which his
colleague had flinched from the agreement to attack, and sacrificed
the certainty of victory. He showed his resentment by acting up to
the terms of his letter written from Plasencia five days before, i.e. by
announcing to Cuesta that, having carried out his pledge to drive the
French from behind the Alberche, he should now refuse to move
forward, unless he were furnished with transport sufficient to make
it certain that the army could reach Madrid without any privations.
He was able to state with perfect truth that he had already been
forced to place his troops on half-rations that very morning: to the
10,000 men of Sherbrooke’s and Mackenzie’s divisions and of
Anson’s light cavalry, he had only been able to issue 5,000 rations of
bread[616]. Nothing, of course, could be found at Talavera, where the
French had been quartered for many days. Victor had only been
maintaining his troops by the aid of biscuit sent down from Madrid,
and by seizing and threshing for himself the small amount of corn
which had been sown in the neighbourhood that spring. Wellesley
was wrong in supposing that the 1st Corps had been supporting
itself with ease from the country-side[617]. He was equally at fault
when he asserted that the ‘Spanish army has plenty to eat.’ Cuesta
was at this moment complaining to the Junta that he was short of
provisions, and that the food which he had brought forward from the
Guadiana was almost exhausted. Meanwhile every exertion was
being made to collect flour and transport from the rear: Wellesley
wrote to O’Donoju that he had at last hopes of securing some
wagons from the Plasencia district within three days, and that ‘in the
meantime he might get something to eat.’ He had some days before
sent orders back even so far as Abrantes, to order up 200
Portuguese carts which had been collected there, and the Central
Junta had informed him that a train for his use had already started
from Andalusia. But ‘there was no very early prospect of relieving
the present distress[618].’
 Cuesta was, as might have been expected, as angry with
Wellesley for refusing to move forward from Talavera, as Wellesley
was with Cuesta for missing the great opportunity of July 23. When
informed that the British army was not about to advance any further,
he announced that he for his part should go on, that Victor was in
full flight, and that he would pursue him to Madrid. ‘In that case’
dryly observed Wellesley, ‘Cuesta will get himself into a scrape; but
any movement by me to his assistance is quite out of the question.
If the enemy discover that we are not with him, he will be beaten, or
must return. The enemy will make this discovery to-day, if he should
risk any attempt upon their rearguard at Santa Ollala[619].’ In reply to
the Captain-General’s declaration that he should press Victor hard,
his colleague only warned him that he would be wiser ‘to secure the
course of the Tagus and open communication with Venegas, while
the measures should be taken to supply the British army with means
of transport[620].’ The Spaniard would not listen to any such advice,
and hurried forward; though he had been for many weeks refusing
to fight the 1st Corps when it lay in Estremadura, he was now
determined to risk a second Medellin. Apparently he was obsessed
by the idea that Victor was in full retreat for Madrid, and would not
make a serious stand. Underlying his sudden energy there was also
some idea that he would disconcert his masters of the Central Junta
by recovering the capital: he had discovered, it would seem, that the
Junta had sent secret orders to Venegas, directing him to take
charge of the city on its reconquest, and giving him authority to
nominate the civil and military officers for its administration. If the
Army of Estremadura seized Madrid, while the Army of La Mancha
was still lingering on the way thither, all these plans would be
frustrated[621].
Accordingly Cuesta pushed on very boldly on the afternoon of the
twenty-fourth, dividing his army into two columns, of which one
marched on Santa Ollala by the high-road to the capital, while the
other moved by Cevolla and Torrijos on the side-road to Toledo. He
was uncertain whether Victor had retired by one or by both of these
routes: if all his corps had taken the former path, the natural
deduction was that he was thinking only of Madrid: if the Toledo
road had also been used, there was reason for concluding that the
Marshal must be intending to join Sebastiani and the 4th Corps, who
might be looked for in that direction. Late in the day the Spanish
general ascertained that the main body of Victor’s army had taken
the latter route: he proceeded to follow it, placing his head quarters
that night at Torrijos, only fifteen miles from Toledo. Next morning
he learnt to his surprise and dismay that he had in front of him not
only the 1st Corps, but also Sebastiani and the King’s reserves from
Madrid: for just at this moment the whole French force in New
Castile had been successfully concentrated, and nearly 50,000 men
were gathered in front of the 33,000 troops of the Army of
Estremadura. Venegas’s diversion had utterly failed to draw off the
4th Corps to the East; the King had come down in haste from
Madrid, and thus the whole plan of campaign which the allied
generals had drawn up had been foiled—partly by the sloth of
Venegas, partly by Cuesta’s inexplicable and perverse refusal to fight
on July 23 upon the line of the Alberche.
 SECTION XVI: CHAPTER V
CONCENTRATION OF THE FRENCH ARMIES: THE
KING TAKES THE OFFENSIVE: COMBATS OF
TORRIJOS AND CASA DE SALINAS

It is now necessary to turn to the French camp, in order to realize

the course of events which had led to the concentration of such a
formidable force in the environs of Toledo. Down to the twenty-
second of July Joseph and his adviser Jourdan had remained in
complete ignorance of the advance of Wellesley upon Plasencia, and
seem to have been perfectly free from any apprehension that Madrid
was in danger. Since their return from their fruitless pursuit of the
army of La Mancha, they had been spending most of their energy in
a controversy with Soult. The Duke of Dalmatia, not content with the
command of the three army corps which Napoleon had put at his
disposal, had been penning elaborate dispatches to the King to
demand that the greater part of the remaining French troops in
Spain should be used to co-operate in his projected campaign
against the English in Portugal. He wrote on July 13 to urge on
Joseph the necessity (1) of drawing large detachments from the
armies of Aragon and Catalonia, in order to form a corps of
observation in the kingdom of Leon to support his own rear; (2) of
placing another strong detachment at Plasencia to cover his flank;
(3) of transferring every regiment that could be spared from Madrid
and New Castile to Salvatierra on the Tormes, just south of
Salamanca, in order to form a reserve close in his rear, which he
might call up, if necessary, to strengthen the 60,000 men whom he
already had in hand. He also demanded that Joseph should send him
at once 200,000 francs to spend on the fortification of Zamora, Toro,
and other places on the Douro, as also 500,000 francs more for the
present expenses of the 2nd, 5th, and 6th Corps. If this were
granted him, together with 2,000,000 rations of flour, and a
battering-train of at least forty-eight heavy guns for the sieges of
Ciudad Rodrigo and Almeida, he thought that he should be in a
position to deliver a serious attack on Northern Portugal, and
ultimately to drive the British army into the sea[622].
On the day upon which the Duke of Dalmatia made these
comprehensive demands upon King Joseph, the British army had
been for ten days in Spain, and was preparing to advance from
Plasencia on Madrid. It was therefore an exquisitely inappropriate
moment at which to demand that the greater part of the King’s
central reserve should be sent off from the capital to the
neighbourhood of Salamanca. There were other parts of Soult’s lists
of requisitions which were equally impracticable. It is clear that
Suchet could not have spared a man from Aragon, and that St. Cyr,
with the siege of Gerona on his hands, would have found it
absolutely impossible to make large detachments from Catalonia.
Even if he and Suchet had been able to send off troops to Leon, they
would have taken months to reach the Galician frontier. The demand
for 700,000 francs in hard cash was also most unpalatable: King
Joseph was at this moment in the direst straits for money: his
brother could send him nothing while the Austrian war was in
progress, and as he was not in proper military possession of any
large district of Spain, he was at this moment in a condition of
hopeless bankruptcy. He confessed to Soult that he was living from
hand to mouth, by the pitiful expedient of melting down and coining
the silver plate in the royal palace at Madrid.
Jourdan therefore replied, in the King’s behalf, to Soult that he
must do his best with the 60,000 men already at his disposition, that
no troops from Catalonia, Aragon, or Madrid could be spared, and
that money could not be found. All that could be given was the
battering-train that had been demanded, 600,000 rations of biscuit,
and an authorization to raise forced contributions in Old Castile. For
the protection of his flanks and his communications the Marshal
must utilize Kellermann’s dragoons and the other unattached troops
in the valley of the Douro, a force which if raised to 12,000 men by
detachments from the 5th or 6th Corps could keep La Romana and
the Galicians in check[623].
It is curious to note how entirely ignorant both Soult and the King
were as to the real dangers of the moment. Soult had drawn up, and
Joseph acceded to[624], a plan for the siege of Ciudad Rodrigo, and
an invasion of Northern Portugal—operations which would take long
weeks of preparation—at the time when Madrid was in imminent
danger from the combined armies of Wellesley, Cuesta, and
Venegas. The Marshal’s plan was perfectly correct from the point of
view of the higher strategy—the main objective of the French was
certainly the British army, and it would have been highly advisable to
invade Northern Portugal with 60,000 men in the front line, and
40,000 in support, if the circumstances of the moment had
permitted it. But these circumstances were hidden alike from Soult
and the King, owing to the impossibility of obtaining accurate
information of the movements of the allies. The fundamental
difficulty of all French operations in the Peninsula was that the
commanders could never discover the whereabouts of the enemy till
he actually came in contact with their outposts. Hence it chanced
that Soult was planning, and Joseph approving, a campaign on the
borders of Northern Portugal, at the precise moment when the
British were on the march for Talavera.
It was actually not until July 22 that the King’s eyes were at last
unsealed. Victor having come into collision with the cavalry of
Wellesley’s advanced guard, sent news to Madrid that the British
army had joined Cuesta, and had reached the Alberche. On the
same day, by a fortunate chance, there also arrived in the capital
another emissary of Soult, with a message much less impracticable
than that which had last been sent. This was General Foy, whom the
Duke of Dalmatia had dispatched on July 19, after receiving very
definite rumours that the British were moving in the valley of the
Tagus, and not approaching Old Castile[625]. The Marshal sent word
that in this case he must of course concert a common plan of
operations with the King, and abandon any immediate action against
Portugal. He suggested that his best plan would be to concentrate
his three corps at Salamanca, and to march against the flank and
rear of the English by way of Bejar and the Puerto de Baños. If the
King could cover Madrid for a time with the 1st and 4th Corps, he
would undertake to present himself in force upon Wellesley’s line of
communications, a move which must infallibly stop the advance of
the allies towards the capital. If they hesitated a moment after his
arrival at Plasencia, they would be caught between two fires, and
might be not merely checked but surrounded and destroyed. Soult
added, however, that he could not move till the 2nd Corps had
received the long-promised provision of artillery which was on its
way from Madrid, and till he had rallied Ney’s troops, who were still
at Astorga, close to the foot of the Galician mountains.
Napoleon, at a later date, criticized this plan severely, declaring
that Soult ought to have marched on Madrid to join the King, and
not on Plasencia. He grounded his objections to the scheme on the
strategical principle that combined operations on external lines
should be avoided. ‘The march of Marshal Soult,’ he wrote, ‘was both
dangerous and useless—dangerous, because the other army might
be beaten (as happened at Talavera) before he could succour it, so
that the safety of all my armies in Spain was compromised: useless,
because the English had nothing to fear; they could get behind the
Tagus in three hours; and whether they crossed at Talavera or at
Almaraz, or anywhere else, they could secure a safe line of retreat
on Badajoz.’ Against this criticism the defence made by both Soult
and King Joseph was that it would have required a much longer time
to bring the three corps from the Douro to Madrid than to Plasencia;
that it would have taken them at least ten days to reach Madrid, and
that during those days the King and his army might have been
beaten and driven out of the capital by the united forces of
Wellesley, Cuesta, and Venegas. It was, of course, impossible to
foresee on July 22 that Wellesley would refuse to pursue Victor
beyond Talavera, or that Venegas would let Sebastiani slip away
from him. Accordingly King Joseph and Jourdan fell in with Soult’s
suggestion, because they thought that he would come sooner into
the field if he marched on Plasencia, and would remove the pressure
of the British army from them at a comparatively early date. As a
matter of fact, he took a much longer time to reach Plasencia than
they had expected: they had hoped that he might be there on July
27, while his vanguard only reached the place on August 1, and his
main body on the second and third[626]. But it seems clear that the
expectation that he would intervene on the earlier date was far too
sanguine. Soult dared not move till his three corps were well closed
up, and since Ney had to come all the way from Astorga, it would
have been impossible in any case to mass the army at Plasencia
much earlier than was actually done. Napoleon’s remark that Soult
could not hope to catch or surround the British army seems more
convincing than his criticism of the march on Plasencia. If the passes
of the Sierra de Gata had been properly held, and prompt news had
been transmitted to Talavera that the French were on the move from
the valley of the Douro, Wellesley would have had ample time to
cover himself, by crossing the Tagus and transferring his army to the
line of operations, Truxillo-Badajoz. The British general always
defended himself by this plea: and complained that those who spoke
of him as being ‘cut off from Portugal,’ by the arrival of Soult at
Plasencia, forgot that he had as good a base at Elvas and Badajoz as
at Abrantes.
But we must not look too far forward into the later stages of the
campaign. It is enough to say that Jourdan and Joseph sent back
Foy to rejoin Soult, on the same day that he had reached Madrid,
bearing the orders that the Marshal was to collect his three corps
with the greatest possible haste, and to march by Salamanca on
Plasencia, where they trusted that he might present himself on the
twenty-seventh or twenty-eighth of the current month. Meanwhile it
was necessary to hold back Cuesta and Wellesley till the Duke of
Dalmatia’s operations in their rear began to produce their effect. The
only possible way of doing this was to concentrate in all haste every
available man in New Castile, and to cover Madrid as long as
possible. This massing of the French forces turned out to be
perfectly feasible, since Venegas had neglected to press in upon
Sebastiani, so that it was possible to withdraw the whole 4th Corps
from in front of him, and to send it to reinforce Victor, without any
immediate danger. Accordingly, the 1st Corps was directed to fall
back from its perilous advanced position on the Alberche, and to
draw near to Toledo: Sebastiani was told to abandon Madridejos and
La Mancha, and to hasten by forced marches toward the same point:
while the King himself resolved to leave Madrid with the slenderest
of garrisons, and to carry the rest of the central reserve to the
general rendezvous. Accordingly, he left only one brigade of
Dessolles’ division, with a few of his untrustworthy Spanish levies, to
hold the capital: the total did not amount to much over 4,000 men,
and General Belliard, the governor of the city, was warned that he
must be prepared to retreat into the Retiro forts, with his troops and
the whole body of the Afrancesados and their families, if anything
untoward should occur. For it was possible that an insurrection might
break out, or that Venegas might succeed in slipping into Madrid by
the roads from the east, or again, that Wilson (whose column had
been heard of at Escalona and was believed to be much larger than
was actually the case), might attempt a coup de main from the west.
Leaving Belliard in this dangerous and responsible position, the King
marched out upon the twenty-third with the remaining brigade of
Dessolles’s division, the infantry and cavalry of his French Guard,
two squadrons of chasseurs and fourteen guns, a force of some
5,800 men[627]. He had reached Navalcarnero, with the intention of
joining Victor on the Alberche, when he received the news that the
Marshal had retired towards Toledo, and was lying at Bargas behind
the Guadarrama river. Here Joseph joined him on the morning of
July 25.
On their concentration a force of 46,000 men was collected,
Victor having brought up 23,000, the King 5,800, and Sebastiani
17,500. The latter had placed four of the six Polish battalions of
Valence’s division in Toledo, and was therefore short by 3,000
bayonets of the total force of his corps. With such a mass of good
troops at their disposition, Joseph, Jourdan, and Victor were all
agreed that it was right to fall upon the Spaniards without delay.
They were astonished to find that the British army was not in their
front, but only Cuesta’s troops. They had expected to see the whole
allied host before them, and were overjoyed to discover that the
Estremadurans alone had pushed forward to Torrijos and Santa
Ollala. Instead, therefore, of being obliged to fight a defensive battle
behind the river Guadarrama, it was in their power to take the
offensive.
This was done without delay: on the morning of July 26 the
French army advanced on Torrijos, with the 1st Corps at the head of
the column. But Cuesta, when once he had discovered the strength
of the force in his front, had resolved to retreat. Victor found
opposed to him only the division of Zayas and two cavalry
regiments, which had been told off to cover the withdrawal of the
Estremaduran army. The Marshal sent out against this rearguard the
chasseurs of Merlin and the dragoons of Latour-Maubourg, who
drove in the Spanish horse, almost exterminating the unfortunate
regiment of Villaviciosa, which, in retiring, chanced to blunder
against the high stone walls of some enclosures from which exit was
difficult[628]. Zayas then went to the rear, and retired towards the
cavalry division of Albuquerque, which Cuesta hastily sent to his
assistance. The French cavalry took some time to re-form for a
second attack, and their infantry was still far off. The Spanish
rearguard therefore, covered by Albuquerque’s horse, had time
enough to fall back on the main body, which was already in full
retreat. Their cavalry then followed, and being not very strenuously
pursued by Merlin and Latour-Maubourg, got off in safety. The whole
army, marching at the best of its speed, and in considerable disorder,
finally reached the Alberche without being caught up by the enemy.
Cuesta found the British divisions of Sherbrooke and Mackenzie
guarding the river: Wellesley had sent them forward when he heard
of the approach of the French, and had placed the former on the
hills above the further side of the bridge, to cover the passage, and
the latter in reserve. He rode out himself to meet the Spanish
general, and begged him to carry his army beyond the Alberche, as
it would be extremely dangerous to be caught with such an obstacle
behind him, and no means of retreat save a long bridge and three
fords. But Cuesta tempted providence by declaring that he should
encamp on the further bank, as his troops were too exhausted to
risk the long defile across the bridge after dark. His sullen anger
against Wellesley for refusing to follow him on the twenty-fourth was
still smouldering in his breast, and the English were convinced that
he remained on the wrong side of the river out of pure perversity,
merely because his colleague pressed him to put himself in safety.
He consented, however, to retreat next morning to the position
which Wellesley had selected in front of Talavera.
The French made no appearance that night, though they might
well have done so, and the Spanish army, bivouacing confusedly in
the narrow slip of flat ground between the heights and the Alberche,
enjoyed undisturbed rest during the hours of darkness. It is
impossible not to marvel at the slackness with which Victor
conducted the pursuit: he had twelve regiments of splendid cavalry
to the front[629], and could undoubtedly have pressed the
Estremadurans hard if he had chosen to do so. Cuesta’s retreating
columns were in such a state of confusion and disorder that a
vigorous assault on their rear might have caused a general
débandade. But after driving in Zayas in the early morning, Victor
moved very slowly, and did not even attempt to roll up
Albuquerque’s cavalry rearguard, though he could have assailed it
with very superior numbers. When taxed with sloth by Marshal
Jourdan, he merely defended himself by saying that the horses were
tired, and that the infantry was still too far to the rear to make it
right for him to begin a combat which might develop into a general
engagement. But it is hard to see that he would have risked
anything by pressing in upon Albuquerque, for if Cuesta had halted
his whole army in order to support his rearguard, there was nothing
to prevent the French cavalry from drawing off, and refusing to close
till the main body of the 1st Corps should come up.
Thanks to Victor’s slackness the Spaniards secured an unmolested
retreat across the Alberche on the following morning. It is said that
Cuesta, in sheer perversity and reluctance to listen to any advice
proffered him by Wellesley, delayed for some hours before he would
retreat, and that when at last he yielded to the pressing solicitations
of his colleague he remarked to his staff ‘that he had made the
Englishman go down on his knees’ before consenting.
All through the morning hours of the twenty-seventh the Army of
Estremadura was pouring across the bridge and the fords, not in the
best order. They had almost all passed, when about noon the French
cavalry began to appear in their front. When the enemy at last
began to press forward in strength, Wellesley directed Sherbrooke’s
and Mackenzie’s divisions to prepare to evacuate their positions on
the eastern bank, which they did as soon as the last of the
Spaniards had got into safety. The first division passed at the bridge,
the third at the fords near the village of Cazalegas: then Sherbrooke
marched by the high-road towards Talavera, while Mackenzie, who
had been told off as the rearguard, remained with Anson’s light
horse near the ruined Casa de Salinas, a mile to the west of the
Alberche.
It may seem strange that Wellesley made no attempt to dispute
the passage of the river, but the ground was hopelessly indefensible.
The left bank (Victor’s old position of July 22) completely commands
the right, the one being high, the other both low and entirely
destitute of artillery positions. Moreover, a great part of the terrain
was thickly strewn with woods and olive plantations, which made it
impossible to obtain any general view of the country-side. They
would have given splendid cover for an army advancing to storm the
heights on the French bank, but were anything but an advantage to
an army on the defensive. For, unable to hold the actual river bank
because of the commanding hills on the further side, such an army
would have been forced to form its line some way from the water,
and the tangled cover down by the brink of the stream would have
given the enemy every facility for pushing troops across, and for
pressing them into the midst of the defender’s position without
exposing them to his fire. Wellington had examined the line of the
Alberche upon the twenty-fourth and twenty-fifth, and had
pronounced it absolutely untenable; ‘no position could be worse,’ he
wrote to O’Donoju[630], but he had discovered one of a very different
kind a little to the rear, and had already settled the way in which it
was to be occupied. It presented so many advantages that even
Cuesta had consented to accept it as a good fighting-ground, and
the Estremaduran army was at this very moment occupied in
arraying itself along that part of the line which had been allotted to
it. Sherbrooke’s division was retiring across the plain to fall into the
section which Wellesley had chosen for it, and Hill’s and Campbell’s
troops were moving to their designated ground. Only Mackenzie and
the light cavalry had yet to be established in their post.
In the act of withdrawing, this division became involved in an
unfortunate combat, which bid fair for a moment to develop into a
disaster. Its two brigades had been halted close to the ruined house
called the Casa de Salinas, in ground covered partly with underwood
and partly with olive groves. The cavalry had been withdrawn to the
rear, as it was impossible to use it for vedettes in such a locality. The
infantry was supposed to have a chain of pickets thrown out in its
front, but it would appear that they must have been badly placed: as
one eye-witness confesses, ‘we were by no means such good
soldiers in those days as succeeding campaigns made us, and
sufficient precautions had not been taken to ascertain what was
passing in the wood[631],’ and between it and the ford below
Cazalegas. French cavalry alone had hitherto been seen, and from
cavalry Mackenzie’s troops were certainly safe in the tangled ground
where they were now lying.
But already Victor’s infantry had reached the front, and its leading
division, that of Lapisse, had forded the Alberche far to the north,
and had entered the woods without being observed by the outlying
pickets of Mackenzie’s left brigade[632]. It had even escaped the
notice of Wellesley himself, who had just mounted the roof of the
ruined Casa de Salinas, the only point in the neighbourhood from
which anything like a general view of the country-side could be
secured. While he was intent on watching the heights above the
Alberche in his front, and the cavalry vedettes descending from
them, the enemy’s infantry was stealing in upon his left.
Lapisse had promptly discovered the line of British outposts, and
had succeeded in drawing out his division in battle order before it
was observed. He had deployed one regiment, the 16th Léger, as a
front line, while the rest of his twelve battalions were coming on in
support.
While, therefore, Wellesley was still unconscious that the enemy
was close upon him, a brisk fire of musketry broke out upon his left
front. It was the French advance driving in the pickets of Donkin’s
brigade. The division had barely time to stand to its arms—some
men are said to have been killed before they had risen from the
ground—and the Commander-in-chief had hardly descended from
the roof and mounted his charger, when the enemy was upon them.
The assault fell upon the whole front of Donkin’s brigade, and on the
left regiment (the 2/31st) of that of Mackenzie himself. So furious
and unexpected was it, that the 87th, 88th, and 31st were all
broken, and driven some way to the rear, losing about eighty
prisoners. It was fortunate that the French advance did not strike
the whole line, but only its left and centre. The 1/45th, which was
just outside the limit of Lapisse’s attack, stood firm, and on it
Wellesley re-formed the 31st, while, a little further to the north, the
half-battalion of the 5/60th also held its ground and served as a
rallying-point for the 87th and 88th. The steadiness of the 1/45th
and 5/60th saved the situation; covered by them the division retired
from the woods and formed up in the plain, where Anson’s light
horsemen came to their aid and guarded their flanks. The French
still pressed furiously forward, sending out two batteries of horse
artillery to gall the retreating columns, but they had done their
worst, and during the hours of the late afternoon Mackenzie’s
infantry fell back slowly and in order to the points of the position
which had been assigned to them. Donkin’s brigade took post in the
second line behind the German Legion, while Mackenzie’s own three
regiments passed through the Guards and formed up in their rear.
Their total loss in the combat of Casa de Salinas had been 440 men
—the French casualties must have been comparatively insignificant—
probably not 100 in all[633].
From the moment when the fray had begun in the woods till
dusk, the noise of battle never stopped, for on arriving in front of
the allied position, the French artillery drew up and commenced a
hot, but not very effective, fire against those of the troops who held
the most advanced stations. As the cannonade continued, the
different regiments were seen hurrying to their battle-posts, for,
although the arrangements had all been made, some brigades, not
expecting a fight till the morrow, had still to take up their allotted
ground.
‘The men, as they formed and faced the enemy, looked pale, but
the officers riding along their line, only two deep, on which all our
hopes depended, observed that they appeared not less tranquil than
determined. In the meanwhile the departing sun showed by his rays
the immense masses moving towards us, and the last glimmering of
the light proved their direction to be across our front, toward the
left. The darkness, only broken in upon by the bursting shells and
the flashes of the French guns, closed quickly upon us, and it was
the opinion of many that the enemy would rest till the morning[634].’
Such, however, was not to be the case: there was to be hard
fighting in front of Talavera before the hour of midnight had arrived.
 SECTION XVI: CHAPTER VI
THE BATTLE OF TALAVERA: THE PRELIMINARY
COMBATS
(JULY 27-28)

The position which Wellesley had selected as offering far better

ground for a defensive battle than any which could be found on the
banks of the Alberche, extends for nearly three miles to the north of
the town of Talavera. It was not a very obvious line to take up, since
only at its northern end does it present any well marked features.
Two-thirds of the position lie in the plain, and are only marked out
by the stony bed of the Portiña, a brook almost dried up in the
summer, which runs from north to south and falls into the Tagus at
Talavera. In the northern part of its course this stream flows at the
bottom of a well-marked ravine, but as it descends towards the town
its bed grows broad and shallow, and ceases to be of any tactical or
topographical importance. Indeed, in this part of the field the
fighting-line of the allies lay across it, and their extreme right wing
was posted upon its further bank.
The town of Talavera, a place of 10,000 souls, which had been a
flourishing industrial centre in the sixteenth century, but had long
sunk into decay, lies in a compact situation on the north bank of the
Tagus. It possesses a dilapidated bridge of forty-five arches, the only
passage across the river between Arzobispo and Toledo. Its site is
perfectly flat, save for a low knoll crowned by the chapel of Nuestra
Señora del Prado, just outside the eastern, or Madrid, gate, and
overlooking the Alameda (public promenade) and the neighbouring
gardens. The place had no suburbs, but was surrounded by a broad
belt of olive groves and enclosures, which extend for a full mile to
the north and east, and hide the houses and walls from the traveller
approaching from either of those directions. When the allies entered
Talavera they found it deserted by most of its inhabitants, who had
fled up into the villages of the Sierra de Toledo during the French
occupation. Many, however, descended to reoccupy their homes
when the enemy departed. Victor’s men had plundered most of the
houses, and turned many of the churches into barracks or stables:
hence the town presented a picture of abject desolation[635].
For a mile and a half beyond the northern wall of Talavera the
ground covered by gardens and olive groves is perfectly flat; it then
commences to rise, and swells up into a long hill, the Cerro de
Medellin. This height runs from east to west, so that its front, and
not the full length of its side, overhangs the Portiña ravine. Its
loftiest point and its steepest face are presented to that declivity,
while to the west and south it has gentle and easily accessible
slopes, sinking gradually down into the plain. This hill, the most
commanding ground in the neighbourhood of Talavera, had been
chosen by Wellesley as the position of his left wing. It formed,
including its lower slopes, about one-third of the line which he had
determined to occupy, the rest of the front lying in the low ground
among the olives and gardens. North of the Cerro de Medellin is a
narrow lateral valley, only half a mile broad, separating this hill from
the main chain of the Sierra de Segurilla, the mountains which form
the watershed between the basin of the Tagus and that of the Tietar.
The British general had intended at first that his position should
extend no further north than the hill, but in the course of the action
he was compelled to lengthen his front, and to post troops both in
the valley and on the mountain spurs beyond it.
By the agreement made with Cuesta, at the conference near the
bridge of the Alberche on the evening of the twenty-sixth, it was
settled that the Spanish army should hold the town of Talavera and
the wooded and enclosed ground for a mile beyond it. The British
had their right among the olive groves, but their centre and left on
the open slopes of the Cerro de Medellin. This order of battle was
the only one which it was possible to adopt. Wellesley had already
discovered that the army of Estremadura could not manœuvre, and
would be much safer behind walls and enclosures than in the open,
and Cuesta had gladly accepted the proposal that he should occupy
this part of the position. Having only a little more than a mile of
front to defend, he was able to provide a double and triple line with
his 32,000 men[636]. His Vanguard and 1st division, under Zayas,
occupied the eastern outskirts of the town, with a battery placed
upon the knoll crowned by the chapel of Nuestra Señora del Prado. A
brigade of cavalry (four regiments) was deployed in the open ground
of the Prado, close to the bank of the Tagus. The 2nd division, that
of Iglesias, held Talavera, whose ancient walls, though imperfect in
many places, were still quite defensible. The 3rd and 4th divisions
(Manglano and Portago) were ranged in a double line among the
gardens and enclosures to the north of the town, as far as a low
hillock called the Pajar de Vergara, where they touched Wellesley’s
left. Behind them were the rest of Cuesta’s cavalry (ten regiments)
and the 5th division (Bassecourt) forming the reserves.
The Spanish position was immensely strong. The front was
completely screened by groves and enclosures occupied by
skirmishers: the first line was drawn up along the slightly sunken
road leading from Talavera to the north, which provided the men
with an excellent parapet and good cover[637]. The second line was
equally well placed behind the Portiña rivulet, which was bordered
by trees along its whole front. The only good artillery position was
that outside the Madrid gate, in front of Zayas’ division, but three
other batteries were planted in the least defective emplacements
that could be found in the front line. The rest of the Spanish guns
were in reserve, in line with Bassecourt and the cavalry.
The northern half of the position had its strong points, but also its
defects. For the first half mile beyond the Spanish left it was still
covered by groves and gardens, and had on its right front the little
eminence of the Pajar de Vergara. On this knoll a redoubt had been
commenced, but no more had been done than to level a space,
eighty yards long and twenty feet broad, on its summit, and to throw
up the excavated earth in front, thus forming a bank three or four
feet high. In this work, indifferently well protected, lay Lawson’s
battery of 3-pounders, the lightest guns of Wellesley’s artillery.
Beside and behind them were the five battalions of the 4th division,
Campbell’s brigade in the front line, Kemmis’s in the second, to the
rear of the Portiña.
On the left of the 4th division the enclosed ground ended, and
cover ceased. Here, forming the British centre, were drawn up the
eight battalions of Sherbrooke’s division, in a single line. The Guards’
brigade, under Henry Campbell, was in perfectly flat level ground,
without shade or cover. Next to them, where there is a gentle ascent
towards the foot of the Cerro de Medellin, were Cameron’s two
battalions; while the two weak brigades of the King’s German
Legion, under Langwerth and Low, continued the front on to the
actual hill, with the Portiña, now flowing in a well-marked ravine, at
their feet[638]. The whole of this part of the British line was bare
rolling ground covered with long dry grass and scattered shrubs of
thyme. There was no cover, and before the Guards’ and Cameron’s
brigades the front was not defined by any strong natural feature. On
the other hand, the terrain on the opposite side of the Portiña was
equally bare, and gave no advantage to an enemy about to attack.
It was otherwise in the portion of the front where the four
German battalions of Langwerth and Low were placed. They had a
steep ravine in front of them, but on the opposite side, as a
compensating disadvantage, the rolling upland swells into a hill
called the Cerro de Cascajal, which, though much less lofty than the
Cerro de Medellin, yet afforded good artillery positions from which
the English slopes could be battered.
Behind Sherbrooke’s troops, as the second line of his centre,
Wellesley had drawn up his 3rd division and all his cavalry. Cotton’s
light dragoons were in the rear of Kemmis’s brigade of the 4th
division. Mackenzie’s three battalions supported the Guards: then
came Anson’s light and Fane’s heavy cavalry, massed on the rising
slope in the rear of Cameron. Lastly Donkin’s brigade, which had
suffered so severely in the combat of Casa de Salinas, lay high up
the hill, directly in the rear of Low’s brigade of the King’s German
Legion.
It only remains to speak of the British left, on the highest part of
the Cerro de Medellin. This section of the front was entrusted to
Hill’s division, which was already encamped upon its reverse slope.
Here lay the strongest point of the position, for the hill is steep, and
well covered in its front by the Portiña, which now flows in a deep
stony ravine. But it was also the part of the British fighting-ground
which was most likely to be assailed, since a quick-eyed enemy could
not help noting that it was the key of the whole—that if the upper
levels of the Cerro de Medellin were lost, the rest of the allied line
could not possibly be maintained. It was therefore the part of the
position which would require the most careful watching, and
Wellesley had told off to it his most capable and experienced
divisional general. But by some miscalculation, on the evening of the
twenty-seventh Hill’s two brigades were not lying on their destined
battle-line, but had halted half a mile behind it—Richard Stewart’s
battalions on the left, Tilson’s on the right flank of the reverse slope.
It is difficult to see with whom the responsibility lay, for Wellesley
was far to the right, engaged in planting Mackenzie’s troops in their
new position behind the centre, while Hill had ridden over towards
Talavera to search for his Commander-in-chief and question him
about details, and returned rather late to give his brigadiers the
exact instruction as to the line they were to take up at nightfall[639].
There were piquets on the crest, and the greater part of the front
slopes were covered by Low’s two battalions of the King’s German
Legion, but the actual summit of the Cerro was not occupied by any
solid force, though the brigades that were intended to hold it lay
only 800 yards to the rear. It was supposed that they would have
ample time to take up their ground in the morning, and no one
dreamt of the possibility of a night attack.
Of the very small force of artillery which accompanied the British
army, we have already seen that Lawson’s light 3-pounder battery
had been placed in the Pajar de Vergara entrenchment. Elliott’s and
Heyse’s were in the centre of the line; the former placed in front of
the Guards, the latter before Langwerth’s brigade of the German
Legion. Rettberg’s heavy 6-pounders were on the Cerro de Medellin,
with Hill’s division: at dusk they had been brought back to its rear
slope and were parked near Richard Stewart’s brigade. Finally
Sillery’s battery was in reserve, between the two lines, somewhere
behind Cameron’s brigade of Sherbrooke’s division. This single unit
was the only artillery reserve of which Wellesley could dispose.
The precise number of British troops in line was 20,194, after
deducting the losses at Casa de Salinas; that of the Spaniards was
within a few hundreds of 32,000. The French, as we have already
seen, had brought a little more than 46,000 men to the field, so that
the allies had a superiority of some 6,000 in mere numbers. If
Wellesley could have exchanged the Army of Estremadura for half
their strength of British bayonets, he might have felt quite
comfortable in his strong position. But his confidence in the value of
his allies, even when firmly planted among walls and groves, was
just about to receive a rude shock.
It was about seven o’clock when the heads of Victor’s columns,
following in the wake of the horse artillery which had been galling
Mackenzie’s retreat, emerged from the woods on to the rolling
plateau facing the allied position. Ruffin appeared on the right, and
occupied the Cascajal hill, opposite the Cerro de Medellin. Villatte
followed, and halted in its rear. More to the left Lapisse, adopting the
same line that had been taken by Mackenzie, halted in front of the
British centre: the corps-cavalry, under Beaumont, was drawn up in
support of him. Latour-Maubourg’s six regiments of dragoons,
further to the south, took ground in front of the Spaniards. The King
and Sebastiani were still far to the rear: their infantry was only just
passing the Alberche, though their advanced cavalry under Merlin
was already pushing forward in the direction of Talavera down the
high-road from Madrid[640].
If Napoleon, or any other general who knew how to make himself
obeyed, had been present with the French army, there would have
been no fighting on the evening of July 27. But King Joseph counted