Towards improving the efficiency of the fuzzy cognitive map classifier

Abstract Fuzzy Cognitive Map (FCM) is a model that combines selected features of fuzzy sets and neural networks. FCM is usually applied as a decision support tool or as a predictive model for time series forecasting. It is less well known as a classifier. To perform the classification, numeric data produced by the FCM must be assigned to class labels. To accomplish this task, we propose a new algorithm for generating thresholds for the discrimination of FCM outcomes. The thresholds resulting from the proposed algorithm are determined after the learning of the FCM, and, then they are applied when classifying new data. The results of the experiments conducted with publicly available data provide evidence that the application of the proposed algorithm leads to improved efficiency of the FCM classifier. Comparative experiments showed that the proposed approach makes the FCM a very competitive alternative to other state-of-the-art classifiers.