A New Approach for Partitional Clustering Using Entropy Notation and Hopfield Network

Abstract

This paper proposes a new clustering algorithm which employs an improved stochastic competitive Hopfield network in order to organize data patterns into natural groups, or clusters, in an unsupervised manner. This Hopfield network uses an entropy based energy function to overcome the problem of insufficient understanding of the data and to obtain the optimal parameters for clustering. Additionally, a chaotic variable is introduced in order to escape from the local minima and gain a better clustering. By minimizing the entropy of each cluster using Hopfield network, we achieve a superior accuracy to that of the best existing algorithms such as optimal competitive Hopfield model, stochastic optimal competitive Hopfield network, k-means and genetic algorithm. The experimental results demonstrate the scalability and robustness of our algorithm over large datasets.