Flexible design of image classification rules using extended Fuzzy Oriented Classifier Evolution

We have previously proposed Fuzzy ORiented Classifier Evolution (FORCE), a method to construct fuzzy classification rules automatically with evolution of a directed graph composed of fuzzy conditions. In this work, we introduce (1) flexible optimization of Membership Functions (MFs) for each condition and (2) comparison operators between two input variables into FORCE. The original FORCE has used fixed MFs because optimizing MFs can decrease interpretability of rules. However optimizing MFs can also improve accuracy and compactness of rules. Hence we introduce flexible optimization of MFs into FORCE to investigate its effectiveness. The comparison operators between two input variables enable rules to consider relationships between two input variables easily, and can also improve performance of FORCE. We apply the proposed model to three different image classification tasks to investigate its performance in comparison with conventional methods.