Improving clustering with hidden Markov models using Bayesian model selection

Abstract

This paper presents a Bayesian clustering methodology that partitions temporal data into homogeneous groups, and constructs state based profiles for each group in the hidden Markov model (HMM) framework. We propose a Bayesian HMM clustering methodology that improves upon existing HMM clustering algorithm by incorporating HMM model size selection into the clustering control structure. Experimental results indicate the effectiveness of our methodology.