Fuzzy clustering algorithms with distance metric learning and entropy regularization

TL;DR

This paper introduces fuzzy clustering algorithms that incorporate distance metric learning and entropy regularization to adaptively determine variable relevance and correlation, improving clustering performance on diverse datasets.

Contribution

It proposes novel fuzzy clustering methods based on Euclidean, City-block, and Mahalanobis distances with entropy regularization, enabling variable relevance and correlation learning.

Findings

Effective variable relevance weighting demonstrated.

Improved clustering accuracy on synthetic and real datasets.

Successful application to noisy image texture segmentation.

Abstract

The clustering methods have been used in a variety of fields such as image processing, data mining, pattern recognition, and statistical analysis. Generally, the clustering algorithms consider all variables equally relevant or not correlated for the clustering task. Nevertheless, in real situations, some variables can be correlated or may be more or less relevant or even irrelevant for this task. This paper proposes partitioning fuzzy clustering algorithms based on Euclidean, City-block and Mahalanobis distances and entropy regularization. These methods are an iterative three steps algorithms which provide a fuzzy partition, a representative for each fuzzy cluster, and the relevance weight of the variables or their correlation by minimizing a suitable objective function. Several experiments on synthetic and real datasets, including its application to noisy image texture segmentation, demonstrate the usefulness of these adaptive clustering methods.