Diabetes Prediction and Classification Using AI Based Techniques

You might also like

Download as pptx, pdf, or txt
Download as pptx, pdf, or txt
You are on page 1of 42

Diabetes Prediction and Classification using AI based Techniques

Master of Technology
in
Department of Electronics and Communication
Digital Electronics and Communication

By:
Guide:
Nayana S
Dr Lakshmi Shrinivasan
1MS19LEC11
Associate Professor
Dept. of ECE
Overview:

• Introduction

• Problem Definition

• Literature Review

• Methodology

• Results and Comparison

• Future work

• Paper submission

• References
Introduction

• Diabetes is one of the fastest growing noncommunicable diseases in the


world.
• Early detection of diabetes prevents hazardous health and saves a life.
• This study helped to predict and diagnose diabetes by selecting the different
methods.
• The hazardous boundary value of the risk factors like GLUCOSE LEVEL,
BLOOD PRESSURE, BMI LEVEL, etc. of PIDD data set is partitioned
using dynamic Membership Function
• Technical Domain – Artificial Intelligence
• Tool: MATLAB
Problem Definition
• The challenge in the diagnosis process is that the clinician has to consider patient’s
symptoms, any possible research mistakes, medical treatment methods, side effects and
also any other diseases with similar symptoms. Adding to these are the enormous
database generated for each patient, with inherent uncertainty in diagnosis lacking
accuracy.

• In the thesis, the following expert systems Type 2 Fuzzy system, Modified Adaptive
Neuro Fuzzy Inference System (ANFIS) and Optimal Decision Tree algorithm are
implemented to determine the probability of diabetes mellitus from the PIMA Indian
Diabetes Database. Performances of these methods are validated in comparison with the
accuracy of the results obtained
Literature Review
1.Review- Lalka, Neeru, and Sushma Jain. "Fuzzy based expert system for
diabetes diagnosis and insulin dosage control" International Conference on
Computing, Communication & Automation, IEEE, 2015
• Fuzzy based expert system to handle dynamics of diabetes diagnosis
• Parameters like BMI, plasma glucose level, min BP & insulin level used to
calculate probability of Type-1 diabetes
• The final output is semantically arranged which depicts different parameters in
terms of the fuzzy numbers like low, medium or high, and the probability of
diagnosis in terms of five such fuzzy numbers like very low, low, medium, high or
very high
• PIDD database from American Diabetes Association is used for diabetes diagnosis
Literature Review (continued)
•PIDD database from American Diabetes Association is used for diabetes diagnosis

• The knowledge base constitutes Type-1 diabetes domain which undergoes


fuzzification and the corresponding fuzzy values are also stored in the knowledge
base.
• Fuzzification performed on knowledge base using trapezoidal membership function.
• Defuzzification performed using centroid method
• Result obtained gives final probability / severity of diagnosis.

Limitations:
• System time response slows down with increase in no of rules
• Lack of adaptation or ability to adjust during a learning process
Literature Review (continued)
2.Kalaiselvi, C. and Nasira, G.M., 2014, February. “A new approach for
diagnosis of diabetes and prediction of cancer using ANFIS.” In 2014 World
Congress on Computing and Communication Technologies (pp. 188-190). IEEE
• Data Mining is one of the most innovative areas of computer science that uses
various statistical techniques, classification, and clustering and pattern
recognition for problems.
• To improve the classification accuracy and to achieve better efficiency a new
approach like Adaptive Neuro Fuzzy Inference System (ANFIS) is proposed.
• The Pima Indian diabetic dataset is used as data set for classification.
Literature Review(Continued)

3. Zhen Sun, Songsen Yu, Yang Zhang. “ An Optimal Decision Tree Model for
Diabetes Diagnosis.” International Conference on Computational Intelligence
and Applications. IEEE, 2019.

• In order to prevent overfitting of the decision tree model, Expectation-


maximization (EM) clustering algorithm is used for data reduction, and then
the data is divided into three data sets.
• The decision tree model is established by different hyperparameters, then the
model with the highest accuracy is selected as the optimal model.
• The model is efficient evaluated by confusion matrix, accuracy, sensitivity
and specificity.
Literature Review(Continued)

4. Priyadarshini L and Lakshmi Shrinivasan .“Design of an ANFIS based Decision


Support System for Diabetes Diagnosis”. International Conference on
Communication and Signal Processing, . IEEE, 2019.

• Adaptive Neuro Fuzzy Inference System (ANFIS) is proposed in the following


research paper.
• This improve the classification accuracy by considering the adaptability factor
through neural networks integrated with fuzzy system.
• The adaptive neuro-fuzzy inference system (ANFIS) is the integration of fuzzy
logic and neural network principles into one frame
Literature Review(Continued)

•Integrate best features of fuzzy systems & neural network


•Advantage of fuzzy control interpolation & adaptability through NN back
propagation
•NN powerful data modelling tool capable of capturing & representing complex
input/ output relationships
• ANFIS- effective tool for tuning membership functions of fuzzy inference
systems
Methodology

General block diagram of Artificial intelligence as an expert system in medical


diagnosis
Methodology
1. “ An Optimal Decision Tree Model for Diabetes Diagnosis.”
• Data mining is a popular means of exploring unknown patterns or
prediction rules. One of the data mining methods is decision tree. A
decision tree is a classification method that uses a tree as a representation.
• In order to prevent overfitting of the decision tree model, Expectation-
maximization (EM) clustering algorithm is used for data reduction, and
then the data is divided into three data sets.
• The decision tree model is established by different hyperparameters, then
the model with the highest accuracy is selected as the optimal model.
• The model is efficient evaluated by confusion matrix, accuracy,
sensitivity and specificity.
• Datasets from Kaggle.

• In order to conduct the research, we used the Pima Indian Diabetes Data
set, which is publicly available from UCI repository
Methodology

Classification Process:

Proposed diabetes classification Partitioned data sets.


process.
Fuzzy Expert System(Type 2)
• Fuzzy system is one of the most efficient qualitative computational method
which can manage large ambiguous dataset to provide precise results.
• The system variables are defined as linguistic terms and fuzzy rules are generated
to model the imprecise aspects of system behaviour.
• Fuzzy logic methodology, assisting the decision system involves three significant
steps- fuzzification, inference engine and defuzzification.
• Type 1 Fuzzy has a limitation that still there some footprint of uncertainty
present, so that can be overcome by using type 2(type reducer).
• Type2 Fuzzy system is characterised by a three-dimensional membership
function.
• Knowledge base comprises of rule base and data source such that rules in the
form of IF-THEN are generated using these data.
Proposed Block Diagram

Crisp Input Inference Crisp Output

Glucose Rule #1 Very Low

De-Fuzzification
Low
Fuzzification

Insulin Rule #2

Aggregation
BMI ||||| Medium

DPF High

Age Rule #n Very high


Methodology

• Knowledge base comprises of rule base and data source such that rules in the
form of IF-THEN are generated using these data
• A total of about 240 rules are generated in the expert system using the PIMA
Indian Diabetes Database (PIDD).
• The outcome of the observation in PIDD is either a ‘0’ indicating negative
diagnosis and ‘1’ indicating positive diagnosis.
• These rules are used to determine the probability of occurrence of diabetes in a
patient based on laboratory tests and displayed symptoms.
• Defuzzification using centroid method to obtain a single crisp output
Methodology
Triangular membership Function was used for the input and Output attributes.
(Modified-ANFIS)

Fig . General block diagram of a neuro-fuzzy hybrid system

• In this algorithm, at first Data Splitting was done and Closed frequency Item set was
calculated.
• Entropy values of the CFI are calculated.
• and centered on the distance the CFIs are clustered using K-medoid clustering algorithm.
Then, these clusters are inputted to the ANFIS for further analysis
Methodology
•The ANFIS structure is described using two fuzzy if-then rules of
model as follows:
• Rule 1: If (x is A1) and
(y is B1) then, f1 = p1 x +q1 y+r1
• Rule 2: If (x is A2) and
(y is B2) then, f2 = p2 x + q2 y+r2
  where,
• x and y are inputs,
• Ai and Bi are fuzzy sets,
• fi are outputs specified by fuzzy rule
within the fuzzy region,
• pi, qi and ri are design parameters obtained
in the training process

 
Methodology

• Confusion Matrix was computed and the performance of the system was
calculated in terms of Accuracy and Sensitivity parameters.

Fig . Structure of a confusion matrix


Simulation Results
1. Decision Tree classification results.

Fig. The Decision Tree Classification Visualization.


Simulation Results

Fig. The probability of occurrence of Diabetes VS Glucose Values.


Simulation Results
Simulation Results
2.Fuzzy based expert system results

Fig. Inputs for the Fuzzy Expert System.


Simulation Results

Fig. Inputs for the Fuzzy IT-2 Expert System.


Simulation Results

Fig. Membership function for BMI input attribute.


Simulation Results

Fig. Rule base of 240 rules generated in MATLAB


Simulation Results

Fig. The probability of outcome value obtained from Fuzzy toolbox(Type 1)


Simulation Results

Fig. Surface viewer(Type 2)


Simulation Results

Fig. Training ANFIS model at epoch 3 using back propagation method.


Simulation Results

Fig. Training ANFIS model at epoch 10 using back propagation


method.
Simulation Results

Fig. Training ANFIS model at epoch 20 using back propagation


method.
Simulation Results

Fig. Training ANFIS model at epoch 20 using hybrid method.


Simulation Results

Fig. Neural network design and training


Simulation Results

Fig. Proposed structure of ANFIS model


Simulation Results

Fig. Proposed structure of ANFIS model


Comparison of Results

Method Accuracy Sensitivity

Optimal Decision 92.8% 93.2%


Tree Algorithm

Modified-ANFIS 97.5% 96.9%

Epoch RMSE MSE Epoch RMSE MSE


20 0.1084 0.01175 20 0.06623 0.004386
50 0.1085 0.01175 50 0.06623 0.004386

Table a.)Back propagation method. Table b) : Hybrid Method

Table : Performance of M-ANFIS Model


Comparison of Results
Method Accuracy Sensitivity
ANFIS 88.73% 79.2%
Optimal Decision 92.8% 93.2%
Tree Algorithm

M-ANFIS 97.5% 96.9%

Method Fuzzy ANFIS Fuzzy 2 M-ANFIS


RMSE 0.45924 0.21964 0.2283 0.06623
MSE 0.21090 0.04824 0.05212 0.004386

Table. Comparison of the present work with the previous work


Future Work

• Apply these algorithms on datasets obtained from diagnosis of other ailments like
cardiac problems, cancer and tuberculosis and validate their classification results.
• These expert systems are to be validated with data from larger number of patients
such that they could be useful in detecting dreadful diseases in early stages.
Paper Submission

[1] Nayana S and Lakshmi Shrinivasan, “An Optimal Decision Tree Model for Diabetes
Prediction and Diagnosis” IEEE 4th International Conference on Computing, power and
Communication Technologies (GUCON) 2021, at CCEC in Bangsar South on
September 24-26, 2021.

[2] Nayana S and Lakshmi Shrinivasan, “Design of a Decision Support System for
Diabetes Diagnosis and Prediction using AI Techniques” IEEE Asian Conference on
Innovation in Technology (ASIANCON) 2021, at Pune on August 28-29 ,2021.
References
1.Lalka, Neeru, and Sushma Jain. "Fuzzy based expert system for diabetes diagnosis and
insulin dosage control" International Conference on Computing, Communication &
Automation, IEEE, 2015.

2.Kalaiselvi, C. and Nasira, G.M., 2014, February. “A new approach for diagnosis of
diabetes and prediction of cancer using ANFIS.” In 2014 World Congress on Computing
and Communication Technologies (pp. 188-190). IEEE

3.Zhen Sun, Songsen Yu, Yang Zhang. “ An Optimal Decision Tree Model for Diabetes
Diagnosis.” International Conference on Computational Intelligence and Applications.
IEEE, 2019.

4.Priyadarshini L and Lakshmi Shrinivasan .“Design of an ANFIS based Decision Support


System for Diabetes Diagnosis”. International Conference on Communication and Signal
Processing, . IEEE, 2019.

4.K. Vidhya and R. Shanmugalakshmi “Modified adaptive neuro‑fuzzy inference


system (M‑ANFIS) based multi‑disease analysis of healthcare Big Data”. Springer 2018.

5. Dongrui Wu a , Jerry M. Mendel “Recommendations on designing practical interval


type-2 fuzzy systems”. Engineering Applications of Artificial Intelligence ,Elsevier 2019.

You might also like