Development of a Reinforcement Learning-based Evolutionary Fuzzy Rule-Based System for diabetes diagnosis.

Area of Science:

• Medical Informatics
• Artificial Intelligence in Healthcare
• Computational Intelligence

Background:

• Early disease diagnosis is crucial for preventing severe complications.
• Existing diabetes diagnostic methods often lack interpretability, hindering clinical trust.
• There is a need for accurate and explainable models in diabetes diagnosis.

Purpose of the Study:

• To develop an interpretable and accurate system for diabetes diagnosis.
• To introduce a Reinforcement Learning-based Evolutionary Fuzzy Rule-Based System (RLEFRBS).
• To enhance the explainability of the diagnostic process.

Main Methods:

• Constructed an initial Rule Base (RB) from numerical data.
• Optimized rules by eliminating redundancies and pruning conditions for simplicity.
• Employed a Genetic Algorithm (GA) for rule subset selection.
• Utilized Reinforcement Learning (RL) for evolutionary tuning of membership functions and weight adjustment.
• Implemented a rule stretching method to handle uncovered instances.

Main Results:

• The RLEFRBS model demonstrated high accuracy in diabetes diagnosis.
• The developed RB was found to be compact and highly interpretable.
• The system effectively learned and optimized diagnostic rules.
• Performance was validated on the Pima Indian Diabetes (PID) and BioSat Diabetes Dataset (BDD).

Conclusions:

• The RLEFRBS presents a promising alternative for diabetes diagnosis due to its accuracy and interpretability.
• The model provides a transparent decision-making process, aiding clinical application.
• This approach addresses the limitations of black-box diagnostic models.