Compact and Interpretable Fuzzy Rule Base Construction for TSK Fuzzy Systems

In high-dimensional scenarios, fuzzy systems often face the challenges of rule explosion and degradation of interpretability. To address these issues, this study proposes an Exploration-Exploitation-based Iterative Algorithm (ExIA) for constructing compact and interpretable fuzzy rule bases. ExIA leverages the feature selection mechanism of decision trees to identify and retain key features and converts the trained decision rules into fuzzy rules. Based on the usage statistics of features in the rules, uncovered-feature exploration trees and important-feature exploitation trees are constructed to generate new rules. In addition, ExIA introduces a rule quality evaluation metric to rank and extract the generated rules. By iteratively constructing exploration and exploitation trees, the candidate rules and their rankings are continuously updated, ultimately converging to a stable and high-quality fuzzy rule base. Experimental results demonstrate that the rule base constructed by ExIA maintains low complexity while effectively preserving the semantic integrity and interpretability of the rules. The Takagi-Sugeno-Kang (TSK) fuzzy system trained on this rule base exhibits superior performance in high-dimensional scenarios.