Advancing Diabetes Diagnosis: The Impact of AI and Machine Learning

Introduction

It is estimated that 415 million people suffer from Diabetes in 2015 and it is predicted that by 2040 over 640 million people will suffer from the disease. When the sugar level(glucose) is high and the body cannot regulate it, such individuals are said to have Diabetes. There are two types of diabetes. Type 1- caused by autoimmune pancreatic β-cell (destruction and characterized by absolute insulin deficiency), and type 2-Associated with insulin resistance and relative insulin deficiency. Some of the symptoms of diabetes are: urinating more than regular, slow healing of wounds, loss of weight and fatigue.

Widely used methods of diagnosing diabetes are A1C or HbA1C test, Fasting Blood Plasma (FBP) test, Fasting Plasma Glucose(FPG) test and Oral Glucose Tolerance Test (OGTT).

If detected early, the disease can be controlled but due to the high increase in the occurrence of the disease, methods are ineffective in detecting the disease early. One outstanding models developed was the Fuzzy Rule-Based System(FRBSs) which have high interpretability, but in this article, we are going to be looking into the Reinforcement Learning-based Evolutionary FuzzyRule-Based System (RLEFRBS) for diabetes diagnosis is proposed tohandle the mentioned challenges with the goal of designing a diagnosis system with high accuracy and interpretability.

Get quality help now

Doctor Jennifer

Verified writer

Proficient in: Science

5 (893)

“ Thank you so much for accepting my assignment the night before it was due. I look forward to working with you moving forward ”

+84 relevant experts are online

Hire writer

The proposed model is composed of a two-step process:

• reduced number of rules and conditions
• using Genetic Algorithm (GA) and ReinforcementLearning (RL) to increase the consistency among the rules

RLEFRBS: Reinforcement Learning-based Evolutionary FuzzyRule-Based System for diabetes diagnosis:

The system focuses on Rule-base (RB). Rule base has four steps:

1. Rule learning: Generating rules using numeric data is the first procedure.
   Get to Know The Price Estimate For Your Paper

   Topic

   Deadline: 10 days left

   Number of pages

   Email Invalid email

   By clicking “Check Writers’ Offers”, you agree to our terms of service and privacy policy. We’ll occasionally send you promo and account related email

   "You must agree to out terms of services and privacy policy"

   Write my paper

   You won’t be charged yet!

2. Rule Pruning : After the first procedure, the number of generated rules increases therefore issues arise with problem space, redundant rules need to be eliminated with the pruning process.
3. Pruning rule antecedents: To improve interpretability, elimination and identification of redundant conditions in the antecedent part poised to make more intelligible rules as much as possible and increase the efficiency of data pruning.
4. Evolutionary rule: The rules resulting from pruning the antecedents are used as candidate rules in the rule-selection process.The important steps are detailed in chromosome encoding and fitness function.

Improving the Rb (Rule Base):

Evolutionary Rule Tuning: Induce superior cooperation among the rules is the aim of tuning and by doing this, variations are brought into the shapes of the membership functions to increase their global interaction. The approach makes use of the advantage of the linguistic 2-tuples representation and simplifies the search space by considering a single parameter for each membership function.

Adjusting Rule weights: it is seen that each rule has its weight, weight adjusting is an approach that keeps the interpretability of the FRBS while increasing the classification accuracy. And in addition to weight adjusting, the parameters of the antecedent membership functions are tuned to some extent. An algorithm was created to achieve this.

Rule Stretching: When some instances are not covered by rules and the number of these occurrences increase, it becomes challenging . Rule generalization or stretching is used by removing a few conditions from the antecedent part. Conditions necessary to meet the query instance is needed. Re-evaluation using Laplace accuracy on training data is encouraged and then the rule with the highest evaluation is used to classify the instance.

Experimental Results

The RLEFRBS model was evaluated using the Pima Indian Diabetes (PID) and BioSat Diabetes Dataset (BDD), showing superior accuracy and interpretability compared to existing models. The MATLAB 2016 platform was utilized for simulation and analysis, demonstrating the model's practical applicability in medical diagnosis.

Data collection: As mentioned earlier, the proposed system was evaluated by using Pima Indian Diabetes Dataset(PIDD) from public data which is available online in UCI data repository.

Accuracy: Accuracy was calculated using the below formula

Accuracy=100∗(correctly classified / (correctly classified+incorrectly classified) )

From the research, SVM had an accuracy of 88.8%, Bayes Net 88.54%, Decision Stump 83.72%, AdaBoostM1

85.68%, and the Proposed method(PM) 90.36%.

The were four parameter which were considered:

• Accuracy: The basis for measuring the quality of the model. In this paper the formula for accuracy after applying the feature selection and k-fold technique is this:

Accuracy =( 𝑇rue positive + 𝑇rue Negative) /((True positive + False Negative) + (False Positive + True Negative).

• Precision: The Precision of a model is the portion of significant events among the recovered events. It is additionally alluded to as a positive prescient worth and its calculated with the below formula

Precision= True Positive/(True positive + False Positive)

• Recall: This can be seen as the portion of relevant occurrences retrieved over the total relevant occurrence and it is also known as sensitivity of the model.

Recall= True Positive/(True Positive- False Negative)

• F-Score: This combines the precision and recall by considering its harmonic mean.

F-Score= 2 * ((Precision*Recall)/(Precision + Recall))

Resalts

Five classification algorithms were used: Support Vector Classifier (SVC), Decision Tree (DT), Random Forest (RF), Logistic Regression (LR) and the Multilayer Perceptron (MLP) on the LDA processed data for k values of 2, 4, 5 and 10 for K-Fold cross-validation. The performance parameters used were the precision, recall, F1 Score and accuracy. After the experiments some of the findings were the highest accuracy 78.7% is achieved by multilayer perceptron, highest recall value of 61.26% is achieved by multilayer perceptron, the highest precision value of 72.45% is achieved by multilayer perceptron classifier and highest value of F1 Score 65.97% is achieved by using the multilayer perceptron classifier. It was observed that the multilayer perceptron classifier had the highest values.

Conclusion

The development of the RLEFRBS and the hybrid machine learning model represents a significant leap forward in the early diagnosis of diabetes. These AI-driven approaches not only promise high accuracy but also ensure interpretability, crucial for clinical applications. The future of medical diagnostics lies in the integration of AI and ML, with continued research focusing on optimizing these models for broader healthcare applications.