Pairwise-Constraints Based Semi-Supervised Fuzzy Clustering with Entropy Regularization

Abstract

Compared with traditional unsupervised clustering, semi-supervised clustering is a more powerful computing method and plays a vital role in pattern analysis and machine learning. The reason for these is that semi-supervised approaches can take advantage of semi-supervised information, which can significantly boost the performance of clustering. Existing methods for semi-supervised fuzzy c-means clustering (FCM) suffer from the following issues: it is generally uneasy to assign the appropriate membership degree value based on traditional entropy regularization using semi-supervised information involved in their objective function. To address this problem, we systematically propose a novel Fuzzy Symmetric Relative Entropy Clustering with Pairwise-Constraints (FSREC-PC) by introducing entropy regularization into the objective function. Moreover, FSREC-PC introduces symmetric relative entropy as a regularized term in its objective function such that its resulting formulas have the clear membership meaning compared with the other semi-supervised FCM algorithms. Further experiments conducted in UCI data show that the proposed clustering algorithm can derive a better performance.