Multivariate Beta Dirichlet Process-based Hidden Markov Models Applied to Medical Applications
Narges Manouchehri, Nizar Bouguila · 2022 IEEE International Conference on Industrial Technology (ICIT) · 2022
Hidden Markov Models (HMMs) are known as one of the most capable statistical tools applied to various applications. Determining the number of hidden states has been an attractive topic in research. In this work, we assume a state space with a nonparametric structure. Moreover, we suppose emission probabilities follow multivariate Beta distribution (MB) which is a flexible and powerful distribution. We present multivariate Beta Dirichlet process mixture as a nonparametric extension of finite mixture model. This elegant structure provides more capability to model in fitting data. Such modifications will empower the classical framework of HMMs. We name our novel clustering method multivariate Beta Dirichlet process-based HMM. To learn our proposed model, we apply variational inference as a compelling approach. We evaluate our model performance by testing it on two medical applications including dementia detection and analyzing colonoscopy images. The results indicate that our proposed model has good potential and could be applied as a promising technique.