Flexible Dirichlet Mixture Model for Multi-modal data Clustering

Abstract

Clustering datasets with complex structures, such as multi-modal properties and asymmetric distributions, presents significant challenges in data analysis. To address these issues, this paper introduces the Flexible Dirichlet Mixture Model (FDMM). The model learning is accomplished through the method of moments and the expectation-maximization (EM) algorithm. Empirical
evaluations across diverse datasets, including unimodal and multi-modal data, demonstrate the model’s superior clustering performance. The results confirm FDMM’s adaptability and effectiveness. We find that the FDMM exhibits superior performance when the underlying data structure is complex, compared to the related-works on multi-modal data clustering.