Comparative Study of Fuzzy Rule-Based Classifiers for Medical Applications

Area of Science:

• Medical Informatics
• Artificial Intelligence
• Computational Biology

Background:

• Machine learning (ML) enhances medical decision support systems (MDSS).
• Objective and accurate clinical decisions are crucial for patient outcomes.
• Fuzzy rule-based systems offer a framework for handling uncertainty in medical data.

Purpose of the Study:

• To compare the performance of 16 fuzzy rule-based ML algorithms on medical datasets.
• To evaluate algorithm efficiency in generating interpretable rules for clinical decision support.
• To identify optimal algorithms for medical data analysis.

Main Methods:

• Applied 16 fuzzy rule-based algorithms to 12 medical datasets and real-world data.
• Evaluated algorithms using Matthews correlation coefficient (MCC), area under the curve (AUC), and accuracy (ACC).
• Analyzed the number and size of generated rules for interpretability.

Main Results:

• Gene expression programming (GPR), repeated incremental pruning to produce error reduction (Ripper), and ordered incremental genetic algorithm (OIGA) showed the best average performance.
• 1R, GPR, and C45Rules-C generated the shortest and most interpretable rules.
• GPR demonstrated a balance between classification performance and rule interpretability.

Conclusions:

• Fuzzy logic and gene expression programming (GPR) offer a valuable approach for medical decision support.
• GPR excels at generating concise and interpretable rules from medical data.
• The study highlights the potential of specific ML algorithms to improve clinical decision-making.