Fully Bayesian Libby-Novick Beta Mixture Model with Feature Selection

In this work, we propose a novel clustering algorithm called Libby-Novick Beta mixture model with feature selection which is developed on a new double bounded distribution. Thanks to its additional shape parameters, this new distribution offers much more flexibility compared to conventional distributions in its family or widely used ones such as Gaussian distribution. We learn our proposed model by fully Bayesian inference, estimate model’s parameter with Markov Chain Monte Carlo technique, and apply Gibbs sampling within Metropolis-Hastings for Monte Carlo simulation. Moreover, we integrated feature selection approach simultaneously within the framework to choose the most informative features for our model. To demonstrate the power and capability of this new unsupervised model, we evaluated it on real medical applications and analyzed pathological images of white blood cells and lung and colon cancer images. The outcomes of our experiment indicates the robustness of our model compared to conventional alternatives.