Evolving Fuzzy System with Multivariable Gaussian Participatory Learning and Recursive Maximum Correntropy - eFCE

This paper suggests a new evolving fuzzy approach called eFCE (evolving Fuzzy with Multivariable Gaussian Participatory Learning and Recursive Maximum Correntropy). The approach uses a single pass learning procedure based on a recursive clustering algorithm with participatory learning and multivariable Gaussian membership functions. The eFCE employs first-order Takagi-Sugeno functional rules that may be added, excluded, merged and/or updated depending on the input data. The rules' antecedent is extracted from the clusters, and the consequent parameters are updated by a recursive algorithm of maximum correntropy. The method to create rules uses a compatibility measure and an arousal index. The compatibility measure employs both Euclidean and Mahalanobis distance. The rules elimination procedure combines age and population to exclude inactive rules. Redundant rules are merged if there is a noticeable overlap between two clusters. The performance of the approach is evaluated using instances of times series forecasting. Computational results and comparisons against alternative state-of-the-art evolving models show that the eFCE has better or comparable performance.