Maximum Likelihood-Based Estimation of Finite Multivariate Libby-Novick Beta Mixture Models in Medical Applications

Abstract

The present study offers a Libby-Novick Beta mixture model, which is based on a generalisation of Beta distribution. As a result of having an additional shape parameter, Libby-Novick Beta distribution provides more flexibility than conventional Beta distribution or other common distributions such as Gaussian distribution. To estimate the parameters of this novel mixture model, we applied maximum likelihood technique. To demonstrate the robustness of our proposed model, we compared it to other alternatives. We tested this novel unsupervised model on three real and publicly available medical datasets. The results indicate that this model has better performance compared to Gaussian mixture model and Beta mixture model.