International Journal of Trend in Scientific Research and Development (IJTSRD)

Volume 8 Issue 3, May-June 2024 Available Online: www.ijtsrd.com e-ISSN: 2456 – 6470

Classification of Lumpy Skin Disease using Principal Component

Analysis (PCA)-based Supervised Machine Learning
Rajesh Kumar1, Dr. Sneha Soni2
1
Research Scholar, 2Head and Professor,
1,2
Department of CSE, SIRTE, Bhopal, Madhya Pradesh, India

ABSTRACT How to cite this paper: Rajesh Kumar |

A viral illness, lumpy skin in cattle is spread by mosquitoes and other Dr. Sneha Soni "Classification of
insects that feed on human blood. Animals that have never been Lumpy Skin Disease using Principal
exposed to the virus are mostly affected by the sickness. Milk, meat, Component Analysis (PCA)-based
and domestic and international livestock commerce are all impacted Supervised Machine
Learning" Published
by cattle lumpy skin disease. Traditional lumpy skin disease in International
diagnosis is exceedingly time-consuming, complicated, and resource- Journal of Trend in
constrained. As a consequence, it is essential to use deep learning Scientific Research
algorithms that can categorize the condition with excellent and Development
performance outcomes. In order to segment and classify diseases (ijtsrd), ISSN: 2456- IJTSRD64954
using deep features, deep learning-based segmentation and 6470, Volume-8 |
classification are suggested. Convolutional neural networks with 10 Issue-3, June 2024, pp.648-653, URL:
layers have been selected for this. The created framework is first www.ijtsrd.com/papers/ijtsrd64954.pdf
trained using data gathered from cattle with Cattle's Lumpy Skin
Disease (CLSD). The skin tone is crucial to identifying the damaged Copyright © 2024 by author (s) and
International Journal of Trend in
region when a disease is represented since the characteristics are
Scientific Research and Development
derived from the input photographs. To do this, a color histogram Journal. This is an
was utilized. A deep pre-trained CNN uses this divided region of Open Access article
altered skin color to extract features. Next, a threshold is used to distributed under the
transform the produced result into a binary format. The classifier for terms of the Creative Commons
classification is PCA-Driven Supervised Machine Learning. The Attribution License (CC BY 4.0)
suggested methodology's classification performance has a 96% (http://creativecommons.org/licenses/by/4.0)
CLSD accuracy rate. We give a comparison with cutting-edge
methodologies to demonstrate the efficacy of the suggested
strategies.
KEYWORDS: CLSD, CNN, PCA-Driven Supervised Machine
Learning, Deep Learning, Transfer Learning

I. INTRODUCTION
Skin illnesses include a wide spectrum of ailments  Preventing the spread of germs, viruses, and other
that affect the skin and may be brought on by disorders
bacterial, viral, fungal, allergy, and parasitic  Assisting you in experiencing various feelings,
infections, as well as by the skin itself. Skin such as pain or temperature
infections are illnesses that impact the skin. These  Producing vitamin D;
illnesses have the potential to result in rashes, itching,  Controlling body temperature
inflammation, and other skin abnormalities. While All disorders that irritate, swell, and irritate the skin
some skin diseases are inherited, others are brought and cause rashes and other changes to the look of the
on by lifestyle choices. Ointments, creams, skin are considered skin issues. A total of 1.79% of
medicines, and dietary modifications are all used as all illnesses in the globe are caused by skin
treatments for various skin disorders.
conditions. The American Academy of Dermatology
The human body is covered by and protected by the Association states. In the US, 1 in 4 people have skin
skin, a sizable organ. It performs a number of problems. Skin conditions may range widely in
different tasks, including: intensity and symptoms, and they can be either
 Holding in liquids and avoiding dehydration transient or permanent, painful or not. Some skin

@ IJTSRD | Unique Paper ID – IJTSRD64954 | Volume – 8 | Issue – 3 | May-June 2024 Page 648
 International Journal of Trend in Scientific Research and Development @ www.ijtsrd.com eISSN: 2456-6470
infections may not be serious, while others may be figure an ascent in LSD reports in Africa and a
fatal. consistent number in Europe. Be that as it may, in
spite of the fact that ARIMA predicts a steady
These anomalies are crucial in the identification of a
number of internal disorders, even though the number of episodes in Asia in 2023-2024, NNAR
majority of skin diseases start in the layers of the conjectures an expansion in such flare-ups. This
study's discoveries add to our developing information
skin. The idea that a person's skin reflects their inside
health has some validity. The skin is often the first on the study of disease transmission of LSD.
organ of the body to exhibit observable symptoms of Azeem as well as others [3], Ongoing worldwide
underlying illness because of its visibility and extension of the uneven skin sickness infection
accessibility. Skin anomalies typically point to beyond endemic sub-Saharan nations in the Center
metabolic, cancerous, and glandular illnesses. East and Asia proposes transboundary transmission.
Skin is subject to pathological alterations of various LSD pandemics have been found in Asia's
kinds, including genetic, inflammatory, benign and Bangladesh, India, China, Nepal, Bhutan, Vietnam,
malignant (neoplastic), endocrine, hormonal, Myanmar, Sri Lanka, Thailand, Malaysia, and Laos
traumatic, and degenerative processes, much like interestingly. The dairy and animals ventures in the
other tissues. The condition of the skin is also space are seriously worried by this. This report sums
impacted by emotions. In many respects, the skin's up information on prior episodes in southern Asia and
response to various illnesses and disorders is different accentuates the danger that LSD postures to adjoining
nations. The various systems and activities expected
from that of other tissues. Extensive skin
inflammation, for instance, may interfere with the to control episodes of this as of late distinguished
disorder in Asia are additionally encouraged.
metabolism of other organs and bodily systems,
resulting in anemia, circulatory collapse, temperature Among others, episodes of Punyapornwithaya [4] and
abnormalities, and disturbances of the water and Knotty skin illness (LSD) impacted cow cultivates all
electrolyte balance in the blood. However, due to the over Thailand in 2021-2022. The LSD pandemic that
skin's powerful healing abilities, extensive harm, such cleared the nation started with this episode.
as thermal burns, may be followed by a significant Therefore, a more full comprehension of LSD the
amount of regrowth of the diseased or wounded study of disease transmission is required. This study
regions, with a disproportionally low degree of set planned to find the spatiotemporal examples of
scarring. LSD episodes in regions with a critical number of
dairy ranches. Information from LSD flare-up
II. BACKGROUND
LSD is a serious transboundary sickness that affects examinations assembled from dairy ranches in Khon
the steers business globally, as indicated by Ayesha Kean area, northern Thailand, were analyzed utilizing
Anwar et al. [2]. The point of this study was to spatio-fleeting models such space-time stage,
Poisson, and Bernoulli models. 133 out of 152 dairy
identify designs and huge change focuses as well as
to foresee the recurrence of LSD pandemic reports in ranches had LSD episodes among May and July
2021. Most of dairy ranches (n = 102) were altogether
Africa, Europe, and Asia. Information from the LSD
pandemic report from the World Association for affected by the LSD episodes in June. The paces of
horribleness and mortality from all crowd assaults
Creature Wellbeing, which traversed the years
were 87, 31, and 0.9%, separately. The groups in the
January 2005 through January 2022, were
review region that were considered to be the most
investigated. Auto-backward moving normal
(ARIMA) and brain network auto-backward (NNAR) probable in view of the consequences of all models
were situated in its northernmost area. The Poisson
models were utilized to figure the quantity of LSD
reports, and paired division was used to pinpoint and space-time change models distinguished 15 and 6
spatio-worldly flare-up bunches, separately, however
genuinely significant information change focuses.
Four crucial defining moments were recognized, one the Bernoulli model just found one group. The most
for every mainland. The period between the third and probable, not entirely settled by those reenactments,
had radii of 1.59, 4.51, and 4.44 km, individually. All
fourth change focuses (2016-2019) in the African
information had the most noteworthy mean number of ranches that were a part of the bunch distinguished by
LSD reports. Gigantic episodes happened somewhere the space-time stage model were likewise recognized
in the range of 2015 and 2017 and related with each by the Poisson model, showing that the pestilence
adjustment of the LSD pandemic in Europe. Asia had zone was distinguished by the two procedures. As per
the most LSD grievances in 2019 after the third the review's outcomes, ranchers who own fields
archived changeover point in 2018. In the following inside a one-kilometer span of where the LSD flare-
up began need to utilize more rigid bug vector control
three years (2022-2024), both ARIMA and NNAR

@ IJTSRD | Unique Paper ID – IJTSRD64954 | Volume – 8 | Issue – 3 | May-June 2024 Page 649
 International Journal of Trend in Scientific Research and Development @ www.ijtsrd.com eISSN: 2456-6470
techniques to check the infection's spread. This study an individual from the Poxviridae family. The
adds to a superior comprehension of the significant side effect of uneven skin sickness is the
spatiotemporal example of LSD groups in the Neethling strain, and different side effects incorporate
pestilence region. The discoveries of this study might a couple of gentle varieties of restricted skin knobs.
help chiefs in thinking up methodologies to forestall These side effects additionally affect the mucous
and control future scourges as well as in focusing on layers of inner organs, like the regenerative and
the appropriation of assets to high-gamble with respiratory frameworks. Creatures with this infection,
regions. like dairy cattle, have for all time harmed skin.
Notwithstanding Mishra [5], Thailand is one of the Contaminated cows produce less milk, are bound to
nations where foot and mouth illness flare-ups have be sterile, have unfortunate development, have fetus
brought about impressive monetary misfortunes. A removals, and even fade away. In this review, an
critical early-cautioning technique that could help the engineering in light of AI is made to distinguish the
affliction. After highlight extraction from pretrained
public authority in laying out a FMD checking and
control program is guaging. The objective of this models like VGG-16, VGG-19, and Beginning v3,
study was to display and gauge the event of FMD this engineering utilizes a few classifiers. We
accordingly characterized the gathered elements
flare-up episodes (n-FMD episodes) consistently in
utilizing our physically developed dataset and
Thailand utilizing time-series techniques like the
occasional autoregressive coordinated moving normal classifiers including kNN, SVM, NB, ANN, and LR
to test the work. With this technique, the cutting-edge
(SARIMA), blunder pattern irregularity (ETS), brain
arrangement had a grouping exactness of 92.5% for
network autoregression (NNAR), geometrical
remarkable smoothing state-space model with Box- the test dataset.
Cox change, ARMA mistakes, Pattern and Occasional III. FEATURE SELECTION
parts (TBATS), and half breed strategies. These The process of choosing a subset of the most
systems were utilized to the month to month n-FMD important characteristics to be utilized in the model-
episodes (n = 1209) between January 2010 and building process is known as feature selection, one of
December 2020. The discoveries showed that the n- the crucial and often used machine learning methods
FMD events had a steady pattern from 2010 to 2020, in data mining. To maximize classification accuracy
however from 2014 to 2020, they were by all with comparatively fewer features, feature selection is
accounts turning out to be more terrible. There is an therefore often thought of as an optimization issue. In
occasional pattern to the pandemic events, with the order to do this, duplicate and unnecessary
yearly pinnacle frequently happening among characteristics from raw datasets are often removed
September and November. The single-method time- via feature selection. Since the 1970s, feature
series models, like TBATS (1,0,0), SARIMA (1,0,1) selection approaches have been extensively used in a
(0,1,1) (12), NNAR (3,1,2)12, and ETS (A,N,A), broad range of domains, including the prediction of
produced the best fit. SARIMA-NNAR and NNAR- protein structure classes, the categorization of tracked
TBATS were the mixture models that played out the neurons, text classification, the detection of auditory
best on the approval datasets. Models that incorporate events, and the classification of gene expression data.
irregularity and a non-direct pattern perform better In order to choose the marker genes for lumpy skin
compared to different models in examination. The disease that influence the classification accuracy,
conjectures featured the rising pattern of n-FMD feature selection approaches are also applied. The
episodes in Thailand, which borders a few countries microarray-based molecular categorization of
with an endemic FMD plague and where cross-line illnesses has made significant strides, but it is still a
traffic in cows is normal. Accordingly, making long way from being used in clinical practice.
compelling control techniques and FMD scourge Numerous feature selection techniques, including the
anticipation methodologies in both Thailand and its Fisher score, have been used to date in the choice of
neighbors is basic. feature genes. One of the most popular supervised
Perone and extra [6], The strength of cows is a feature selection techniques is the Fisher score. The
method we'll use delivers, in decreasing order, the
significant worry for people in general. At the point
when sicknesses in creatures are distinguished in the rankings of the variables based on the fisher score.
The variables may then be chosen based on the
beginning phases, they may at times be dealt with.
The livestock business might experience enormous situation.
monetary misfortunes in the event that uneven skin IV. METHODOLOGY
sickness isn't effectively treated. The primary driver In this work, we used a transfer learning strategy to
of this disease is the uneven skin infection, which is classify lumpy skin diseases using the trained

@ IJTSRD | Unique Paper ID – IJTSRD64954 | Volume – 8 | Issue – 3 | May-June 2024 Page 650
 International Journal of Trend in Scientific Research and Development @ www.ijtsrd.com eISSN: 2456-6470
supervised learning model. The output of the pre- X_test_pca \- Apply PCA_MODEL to convert
trained model was flattened and supplied to the X_test
classifier using picture data alone for both binary and
Step 3. Model Training
multi-class classification of LSDs. The classifier then  Set up a supervised learning model (like Logistic
classifies the LSD using the concatenation of both the Regression) from scratch.
picture data and the cattle information. A dense layer MODEL \- Set up a supervised learning model,
with 128 neurons was added on top of the pre-trained such as Logistic Regression.
supervised learning model since the picture data it
produces is greater than the patient information  Use the associated labels ({y_train}) and the
feature. As a result, the model's output image features shortened training set ({X_train_pca}) to train the
are reduced to 128 and the classifier's two input model.
types—one-hot encoded patient information features fit MODEL for both y_train and X_train_pca
and image features—are balanced. In order to address Step 4. Model Evaluation
the dataset's class imbalance, a weighted loss function  Forecast both the test set ({X_test_pca}) and the
based on label frequency was used, which gives fewer training set ({X_train_pca}).
represented classes greater weight. Sure, here is the y_train_pred <- forecast on X_train_pca using
pseudo code for a supervised machine learning MODEL
algorithm using PCA for dimensionality reduction: y_test_pred <- forecast on X_test_pca using
Step 1. Data Preparation MODEL
After loading the dataset, divide it into labels ({y})  Determine the model's accuracy using both the
and features ({X}). training and test sets.
DATASET<- import the dataset testing_accuracy \- use y_test and y_test_pred to
Divide the data into test and training sets. determine accuracy; training_accuracy \- use
y_train and y_train_pred to calculate accuracy
X, y <- divide the dataset into labels and features
divide X and Y into training and test sets (e.g., 80%  Create and print the test set prediction
train, 20% test) X_train, X_test, y_train, y_test classification report and confusion matrix.

Step 2. Applying PCA Classification report generated by utilizing y_test and

 Set the required number of components y_test_pred print CONFUSION_MATRIX {-
({n_components}) at the beginning of the PCA compute confusion matrix using y_test and
model. y_test_pred The training accuracy print is
PCA_MODEL <- Use n_components to initialize "training_accuracy:". The print "Test Accuracy:",
PCA test_accuracy The print "Confusion Matrix:",
CONFUSION_MATRIX
 To get the reduced training set ({X_train_pca}), CLASSIFICATION_REPORT, "Classification
fit the PCA model on the training set ({X_train}) Report:"
and transform it.
X_train_pca \- fit X_train to PCA_MODEL and The general procedure for fusing PCA with a
convert X_train supervised machine learning method is described in
this pseudo code. Modify the particular stages
 To get the reduced test set ({X_test_pca}), according on your dataset and needs.
transform the test set ({X_test}) using the fitted
PCA model.

V. RESULTS AND ANALYSIS

Following the pre-processing of the data in this experiment, the top ten features of the dataset are chosen using
the features selection technique, and then well-known supervised and unsupervised learning methods are put into
practice. Support Vector Machine (SVM), XGBoost, and Extreme Learning Machine (ELM) are used in
machine learning. The suggested proposed model belongs to the Efficient CNN subcategory.

@ IJTSRD | Unique Paper ID – IJTSRD64954 | Volume – 8 | Issue – 3 | May-June 2024 Page 651
 International Journal of Trend in Scientific Research and Development @ www.ijtsrd.com eISSN: 2456-6470

Figure 2: Instances in Dataset

Table 1: Experiment Results of Machine Learning Models
Method Sensitivity (%) Specificity (%) Precision (%) Accuracy (%)
C-SVM 88.89 89.98 989.51 89.08
C-KNN 87.80 88.00 88.10 87.82
Q-SVM 89.01 89.20 89.00 89.03
ESD 89.00 89.89 90.00 89.80
M-SVM 89.00 89.00 90 89.11
XGBoost 90.01 89.00 89.25 89.92
ELM 90.01 90.05 90.19 90.06
Table 1 displays the experiment results for various machine learning methods, whereas Table 1 displays the
experiment results for various machine learning algorithms, including the suggested model. The comparison
between the suggested model and the state-of-the-art model is shown in Table 2.
Table 2: Experiment Results of Different Machine Learning Algorithms including the Proposed
Model
Method Sensitivity (%) Specificity (%) Precision (%) Accuracy (%)
ESD 89.00 89.89 90.00 89.80
M-SVM 89.00 89.00 90 89.11
XGBoost 90.01 89.00 89.25 89.92
ELM 90.01 90.05 90.19 90.06
Proposed 91.31 90.84 92.14 94.11
In this part, we compare the outcomes produced by our suggested simulation model to those produced by earlier
suggested models. Table 4.3 displays our findings and contrasts them with findings from other models.

Figure 3: Performance Analysis of Different Models

@ IJTSRD | Unique Paper ID – IJTSRD64954 | Volume – 8 | Issue – 3 | May-June 2024 Page 652
 International Journal of Trend in Scientific Research and Development @ www.ijtsrd.com eISSN: 2456-6470
VI. CONCLUSIONS northeastern Thailand. Front. Vet. Sci. (2022),
This study offered a segmentation and classification 9, 957306.
model for the lumpy skin condition of cattle. A
[5] Punyapornwithaya, V.; Mishra, P.; Sansamur,
transfer learning approach using MobiltNetV2 was C.; Pfeiffer, D.; Arjkumpa, O.; Prakotcheo, R.;
detailed in the framework. On the well-known Damrongwatanapokin, T.; Jam-pachaisri, K.
datasets for the lumpy skin disease in cattle, the Time-Series Analysis for the Number of Foot
suggested technique was assessed. Different kinds of and Mouth Disease Outbreak Episodes in Cattle
supervised and unsupervised learning approaches are Farms in Thailand Using Data from 2010–
analyzed using F1-Measure, recall, precision, and 2020. Viruses (2022), 14, 1367.
recall, which are briefly addressed. The Fisher score
approach was used to examine and pre-process the [6] Perone, G. Comparison of ARIMA, ETS,
LSD after it was obtained from the Kaggle library. NNAR, TBATS and hybrid models to forecast
This method minimizes the amount of features in the the second wave of COVID-19 hospitalizations
dataset and prevents the over-fitting issue. On the pre- in Italy. Eur. J. Health Econ. (2022), 23, 917–
processed dataset, supervised and unsupervised 940.
machine learning methods are used. The ensemble [7] Maw, M.T.; Khin, M.M.; Hadrill, D.; Meki,
model outperforms all other models including the I.K.; Settypalli, T.B.K.; Kyin, M.M.; Myint,
state-of-the-art model when the performance of all the W.W.; Thein, W.Z.; Aye, O.; Palamara, E. First
methods is compared. Report of Lumpy Skin Disease in Myanmar and
The following are the thesis's future focuses: Molecular Analysis of the Field Virus Isolates.
 To provide a thorough analysis of Deep Learning Microorganisms (2022), 10, 897.
algorithms using a real-time dataset to offer a [8] Singhla, T.; Boonsri, K.; Kreausukon, K.;
better solution for the intrusion detection system. Modethed, W.; Pringproa, K.; Sthitmatee, N.;
 Looking at novel pre-processing methods that Punyapornwithaya, V.; Vinitchaikul, P.
might increase model accuracy. Molecular Characterization and Phylogenetic
Analysis of Lumpy Skin Disease Virus
 The performance of the LSD detection system Collected from Outbreaks in Northern Thailand
may be improved by using deep learning in (2021). Vet. Sci. 2022, 9, 194.
methods.
[9] Khalil, M.I.; Sarker, M.F.R.; Hasib, F.Y.;
REFERENCES Chowdhury, S. Outbreak investigation of
[1] Musa Genemo, “Detecting high-risk Area for lumpy skin disease in dairy farms at Barishal,
Lumpy Skin Disease in Cattle Using Deep Bangladesh. Turk. J. Agric. Food Sci. Technol.
Learning Feature”, Advances in Artificial (2021), 9, 205–209.
Intelligence Research (AAIR), Vol. 3 (No. 1),
pp. 27-35, ISSN 2757-7422, (2023). [10] Lu, G.; Xie, J.; Luo, J.; Shao, R.; Jia, K.; Li, S.
Lumpy skin disease outbreaks in China, since 3
[2] Ayesha Anwar, Kannika Na-Lampang, Narin August 2019. Transbound. Emerg. Dis. (2021),
and Veerasak, “Lumpy Skin Disease Outbreaks 68, 216–219.
in Africa, Europe, and Asia (2005–2022):
Multiple Change Point Analysis and Time [11] Chibssa, T.R.; Sombo, M.; Lichoti, J.K.; Adam,
Series Forecast”, MDPI Journal, (2022). T.I.B.; Liu, Y.; Elraouf, Y.A.; Grabherr, R.;
Settypalli, T.B.K.; Berguido, F.J.; Loitsch, A.
[3] Azeem, S.; Sharma, B.; Shabir, S.; Akbar, H.; Molecular analysis of East African lumpy skin
Venter, E., “Lumpy skin disease is expanding disease viruses reveals a mixed isolate with
its geographic range: A challenge for Asian features of both vaccine and field isolates.
livestock management and food security”. Vet. Microorganisms (2021), 9, 1142.
J. (2022), 279, 105785.
[12] Fatemeh Namazi and Azizollah Khodakaram
[4] Punyapornwithaya, V.; Seesupa, S.; Tafti, “Lumpy skin disease, an emerging
Phuykhamsingha, S.; Arjkumpa, O.; Sansamur, transboundary viral disease: A review”, Vet
C.; Jarassaeng, C. Spatio-temporal patterns of Med Sci. (2021) May; 7(3): 888–896.
lumpy skin disease outbreaks in dairy farms in

@ IJTSRD | Unique Paper ID – IJTSRD64954 | Volume – 8 | Issue – 3 | May-June 2024 Page 653