A resonance correlation network with adaptive fuzzy leader clustering

Cluster analysis is a significant area of research in pattern recognition. Determining the optimal number of clusters in any real data set remains a difficult problem. The paper develops a new neural network model with the combined advantages of self-organization and no sequential search (as in the resonance correlation network) with more stable, fewer and better clusters (as in the adaptive fuzzy leader clustering network). This new model is the Adaptive Fuzzy Leader Clustering Resonance Correlation Network (AFLCRCN). It adaptively clusters continuous-valued data into classes without a priori knowledge of the entire data set or ifs number of clusters. AFLCRCN incorporates the fuzzy K-means learning rule used in the AFLC network into the RCN control structure. It has a modular design that allows metric replacement for improved performance in a specific problem. Applications for the model include classification, feature extraction, and pattern recognition