Simultaneous Clustering and Feature Ranking by Competitive Repetition Suppression Learning with Application to Gene Data Analysis

The paper presents feature-wise Competitive Repetition-suppression (CoRe) clustering, a novel unsupervised algorithm that deals with the automatic determination of the unknown cluster number and simultaneous feature ranking. The proposed model addresses the limitations of the original CoRe learning algorithm when dealing with high dimensional data, extending the repetition suppression competition on a feature-wise basis. The effectiveness of the approach is tested on gene expression data from DNA microarrays: the results show that the feature-wise CoRe clustering algorithm is able to detect the known data partitioning in a completely unsupervised fashion. Moreover, it simultaneously develops a gene ranking that is consistent with the state-of-the-art list of gene relevance for the selected benchmark datasets.