ANN classifier using MATLAB

Presented By
U.Ashadevi
Research Scholar (FT-PhD)
Department of Instrumentation Engineering
MIT Campus, Anna University

13/02/2025 Workshop on "Machine learning and Deep learning for biomed 1

ical applications"
 Organization of the Presentation
• Introduction
• Data resources
• Data analysis
• Classification without data preprocessing
1. ANN architecture
2. Training parameters
3. Performance metrics
 Classification with data preprocessing
1. Multicollinearity analysis
2. Normalization
3. Sensitivity analysis
4. Denormalization
13/02/2025 Workshop on "Machine learning and Deep learning for biomed 2
ical applications"
 Introduction
 An artificial neural network (ANN) is a tool that can be utilized to recognize cancer effectively.
Nowadays, the risk of cancer is increasing dramatically all over the world. Detecting cancer is very
difficult due to a lack of data. Proper data are essential for detecting cancer accurately. Cancer
classification has been carried out by many researchers, but there is still a need to improve classification
accuracy.

13/02/2025 Workshop on "Machine learning and Deep learning for biomed 3

ical applications"
 Cont.,
 Classifying cancers as benign or malignant based on biopsy characteristics is a common
application of machine learning, including techniques like ANN. Biopsies provide crucial
information about tissue samples, and using this data to differentiate between benign (non-
cancerous) and malignant (cancerous) conditions is vital in diagnosis and treatment planning.

 Accurate classification of cancers based on biopsy characteristics is fundamental for providing

patients with the best possible care, optimizing healthcare resources, and advancing research
efforts toward more effective cancer management strategies.

 Pattern recognition is the process of training a neural network to assign

the correct target classes to a set of input patterns. Once trained the
network can be used to classify patterns.
13/02/2025 Workshop on "Machine learning and Deep learning for biomed 4
ical applications"
 Data resources
 Data is available from the UCI Machine Learning Repository .

 http://mlearn.ics.uci.edu/MLRepository.html

 Murphy,P.M., Aha, D.W. (1994). UCI Repository of machine learning databases Irvine, CA: University of
California, Department of Information and Computer Science.

 This dataset can be used to design a neural network that classifies cancers as either benign or malignant
depending on the characteristics of sample biopsies.

 CancerInputs - a 9x699 matrix defining nine attributes of 699 biopsies.

 1. Clump thickness
2. Uniformity of cell size
3. Uniformity of cell shape

13/02/2025 Workshop on "Machine learning and Deep learning for biomed 5

ical applications"
 Cont.,
4. Marginal Adhesion
5. Single epithelial cell size
6. Bare nuclei
7. Bland chomatin
8. Normal nucleoli
9. Mitoses

 CancerTargets - a 2x699 matrix where each column indicates a correct category with a one in either element 1
or element 2.
1. Benign
2. Malignant

 Inputs 'cancerInputs' is a 9x699 matrix, representing static data: 699 samples of 9 elements.

 Targets 'cancerTargets' is a 2x699 matrix, representing static data: 699 samples of 2 elements.

13/02/2025 Workshop on "Machine learning and Deep learning for biomed 6

ical applications"
 Data analysis

13/02/2025 Workshop on "Machine learning and Deep learning for biomed 7

ical applications"
 Cont.,

13/02/2025 Workshop on "Machine learning and Deep learning for biomed 8

ical applications"
 Classification without data preprocessing
1.ANN architecture and Training parameters

Training parameters Values

Neural network model FF
Input nodes 9
Hidden layer 2
Hidden layer neurons 10
Output layer neurons 2
Output nodes 2
Training network algorithm Scaled
conjucate
BP
Data division(Training:Testing:Validation) 70:15:15
Training Function trainscg
Hidden neuron and output neuron Sigmoid,
softmax
Objective function Cross
entropy
No of epochs 1000

13/02/2025 Workshop on "Machine learning and Deep learning for biomed 9

ical applications"
 3.Performance metrics

13/02/2025 Workshop on "Machine learning and Deep learning for biomed 10

ical applications"
Classification performance evaluation
Metrics Training Testing Validation All

confusion confusion confusion confusion

matrix matrix matrix matrix

Accuracy 96.72 98.09 95.23 96.70

Precision 98.38 97.26 97.22 97.80

Specificity 97.12 94.11 90.90 95.85

Sensitivity 96.50 100 97.22 97.16

13/02/2025 Workshop on "Machine learning and Deep learning for biomed 11

ical applications"
 Cont.,
Metrics Training Testing Validation All

confusion confusion confusion confusion

matrix matrix matrix matrix

Accuracy 96.72 98.09 95.23 96.70

Precision 98.38 97.26 97.22 97.80

Specificity 97.12 94.11 90.90 95.85

Sensitivity 96.50 100 97.22 97.16

13/02/2025 Workshop on "Machine learning and Deep learning for biomed 12

ical applications"
 Classification with data preprocessing

13/02/2025 Workshop on "Machine learning and Deep learning for biomed 13

ical applications"
 1.Multicollinearity analysis

(Variance Inflation Factor (VIF))

13/02/2025 Workshop on "Machine learning and Deep learning for biomed 14

ical applications"
 Variance Inflation Factor(VIF)
VIF is majorly used to measure the severity of multicollinearity in a multiple regression model. It is
usually measured as the variance of a model with multiple terms divided by the variance of a model
with one term alone.

13/02/2025 Workshop on "Machine learning and Deep learning for biomed 15

ical applications"
13/02/2025 Workshop on "Machine learning and Deep learning for biomed 16
ical applications"
 Cont.,

13/02/2025 Workshop on "Machine learning and Deep learning for biomed 17

ical applications"
 2.Normalization (or)Standardization
Normalizing data is a valuable pre-processing step before training as it aligns the data distribution to
centre the mean of all data points around zero, as it can enhance the training efficiency and
accelerate network convergence.

Where is the input normalized to lie between 0 and 1, is the original value of the feature , is the
minimum value of the feature and , is the maximum value of the feature. This process scales all
values between 0 and 1.

13/02/2025 Workshop on "Machine learning and Deep learning for biomed 18

ical applications"
 Feature Scaling/Normalization
using MATLAB

Workshop on "Machine learning and Deep learning for

13/02/2025 biomedical applications" 19
 Min-Max Normalization
• % Assuming your data is stored in the 'data' variable, where each row represents a data point and
each column represents a feature.

• % Define the minimum and maximum values for normalization

• min_val = min(data);

• max_val = max(data);

• % Perform min-max normalization

• normalized_data = (data - min_val) ./ (max_val - min_val);

• % Display the normalized data

• disp('Normalized Data (Min-Max):');

•13/02/2025
disp(normalized_data); Workshop on "Machine learning and Deep learning for biomed 20
ical applications"
 Cont.,

13/02/2025 Workshop on "Machine learning and Deep learning for biomed 21

ical applications"
 3.Sensitivity analysis
Sensitivity analysis was performed by computing the correlation coefficients between the model
inputs and outputs.

13/02/2025 Workshop on "Machine learning and Deep learning for biomed 22

ical applications"
 Classification performance evaluation
Metrics Training Testing Validation All

confusion confusion confusion confusion

matrix matrix matrix matrix

Accuracy 97.54 97.14 98.09 97.56

Precision 99.36 100 98.38 99.32

Specificity 98.79 100 97.67 98.75

Sensitivity 96.90 95.89 98.38 96.94

13/02/2025 Workshop on "Machine learning and Deep learning for biomed 23

ical applications"
 Cont.,
Metrics Training Testing Validation All

confusion confusion confusion confusion

matrix matrix matrix matrix

Accuracy 97.54 97.14 98.09 97.56

Precision 99.36 100 98.38 99.32

Specificity 98.79 100 97.67 98.75

Sensitivity 96.90 95.89 98.38 96.94

13/02/2025 Workshop on "Machine learning and Deep learning for biomed 24

ical applications"
 4.Denormalization

• % Denormalize the output classification using the min-max normalization formula

• denormalized_predictions = normalized_classification.* (max_val - min_val) + min_val;

• % Display the denormalized classification

• disp('Denormalized classification :');

• disp(denormalized_ classification);

13/02/2025 Workshop on "Machine learning and Deep learning for biomed 25

ical applications"
13/02/2025 Workshop on "Machine learning and Deep learning for biomed 26
ical applications"