Constructing Interpretable Genetic Fuzzy Rule-Based System for Breast Cancer Diagnostic

This paper shows how a subset of features can be selected for designing interpretable fuzzy rule-based system. This method consists of two phases: feature subset selection based on Michigan Learning approach and Training fuzzy rule-based system using the selected subset from the first phase. First, a number of independent fuzzy rule-based systems are trained using genetic operations, and then the dominated rules of each trained system with the highest fitness values are selected. From the selected rules, a pre-specified number of features are chosen with the highest frequency. In the second phase, a fuzzy rule-based system is trained based on the selected features from the previous phase. Experiments shows the two-phase method feature reduction based on the ldquocollective thoughtrdquo can achieve promising classification accuracy and performance in Breast Cancer Diagnostic Wisconsin data set.