Bayesian Model and Feature Selection in Asymmetric Generalized Gaussian Mixtures

This paper addresses an advanced unsupervised Bayesian framework for learning a finite mixture of asymmetric generalized Gaussian distributions (AGGD) using Markov Chain Monte Carlo (MCMC), incorporating Metropolis-Hastings within Gibbs sampling for estimating mixture parameters and reversible jump MCMC (RJMCMC) to dynamically select the number of model components. To enhance learning, we developed feature selection methods that assign weights to features based on their significance. Numerous experiments are carried out to evaluate the model’s performance in a variety of real-world applications. The performance of our model is compared to other benchmark models in order to determine the significance of proposed learning approach.