Hybrid Grid Partitioning and Rough Set-based Fuzzy Rule Generation for Enhanced Accuracy and Interpretability in Dataset Classification

Abstract

Achieving a balance between accuracy and interpretability is a critical challenge in dataset classification. Traditional methods often prioritize accuracy but lack interpretability, making it difficult to comprehend and trust the classification decisions. In this research, we propose a hybrid approach that combines grid partitioning, rough set-based attribute reduction, and fuzzy rule generation to enhance both the accuracy and interpretability in dataset classification. The grid partitioning technique discretizes continuous attribute values into intervals, reducing complexity and improving accuracy. The rough set-based attribute reduction identifies the most relevant attributes, reducing noise and irrelevant information. Fuzzy rule generation generates human-understandable rules based on the refined attribute values and rough sets, providing explanations for classification decisions. By integrating these techniques, our approach strikes a balance between accuracy and interpretability. The weight parameter in the objective function allows customization according to specific requirements. Experimental evaluations demonstrate the effectiveness of the proposed approach in achieving enhanced accuracy and interpretability. The hybrid approach has potential applications in domains where transparent and understandable classification models are crucial, such as medical diagnosis or decision support systems. Further research can explore parameter optimization and evaluate the approach on diverse datasets to enhance its applicability and robustness.