Partitioning Features for Model-Based Clustering Using Reversible Jump MCMC Technique

摘要

In many cluster analysis applications, data can be composed of a number of feature subsets where each is represented by a number of diverse mixture model-based clusters. However, in most feature selection algorithms, this kind of cluster structure has been less interesting because they accounted for discovery of a single informative feature subset for clustering. In this study, we attempt to reveal a feature partition comprising multiple feature subsets, with each represented by a mixture model-based cluster. Searching for the desired feature partition is performed by utilizing a local search algorithm based on a reversible jump Markov Chain Monte Carlo technique.