Classification and feature transformation with Fuzzy Cognitive Maps

Fuzzy Cognitive Maps (FCMs) are considered a soft computing technique combining elements of fuzzy logic and recurrent neural networks. They found multiple application in such domains as modeling system behavior, prediction time series, decision making process control. Less attention, however, has been turned towards using them pattern classification. In this work we propose an FCM based classifier with fully connected map structure. contrast to methods that expect reaching steady state during reasoning, chose execute few iterations (steps) before collecting output labels. Weights were learned gradient algorithm logloss or cross-entropy used the cost function. Our primary goal was verify, whether design would result descent general purpose classifier, performance comparable off shelf classical methods. As preliminary results promising, investigated hypothesis $d$-step can be attributed fact previous $d-1$ steps it transforms feature space by grouping observations belonging given class, so they became more compact separable. To verify calculated three clustering scores for transformed space. We also evaluated pipelines built from FCM-based data transformer followed classification algorithm. The standard statistical analyzes confirmed both its capability improve data. supporting prototype software implemented Python TensorFlow library.