Pairwise constraints cross entropy fuzzy clustering algorithm based on manifold learning and feature selection

Abstract

In weakly supervised learning, it is difficult for us to utilize pairwise constraints information in feature selection. In order to solve the problem, we propose Pairwise constraints cross entropy fuzzy clustering algorithm based on manifold learning and feature selection (FCPC-LEFS). There are four phases in our approach: 1) Generate pseudo label; 2) Dimension reduction by Laplacian Eigenmaps; 3) Feature increment and selection; 4) Cross-Entropy semi-Supervised Clustering Based on Pairwise Constraints. We apply our approach to three UCI datasets and a COVID19-CT image dataset. Experiments show that our manifold learning and feature selection method are able to increase improve the clustering performance.