Practice on Classification using Gaussian Mixture Model

This project centers on the investigation of appl-ying Gaussian Mixture Model (GMM) to supervised learning based on the Maximum Lik-elihood (ML) estimation using Expectation Maximization (EM). As learnt, the statistical modeling methods manipulate probabilities dire-ctly, thus giving more sophisticated description over the actual world with its disadvantage of the expensive computational complexity. Yet, it is still potential for its hardly use in the field of supervised learning. Based on the model, some modifications are conducted from the classical GMM, thus applying the models to the supervised learning. Two strategies out of the analysis of GMM’s characteristics are put forward and experimented based on some of the weka file from UCI dataset. In this project, it is demanded to implement the basic computations of Gaussian mixture and EM; thus, apart from the understanding of these algorithms, the implementation details offered much potential improvement to deal with, thus leaving a lot to explore further.