Abstract

In this thesis, we propose the consideration of Aitchison’s distance in K-means clustering algorithm. It has been used for initialization of Dirichlet and generalized Dirichlet mixture models. This activity is then followed by that of estimating model parameters using Expectation-Maximization algorithm. This method has been further exploited by using it for intrusion detection where we statistically analyze entire NSL-KDD data-set. In addition, we present an unsupervised learning algorithm for finite mixture models with the integration of spatial information using Markov random field (MRF). The mixture model is based on Dirichlet and generalized Dirichlet distributions. This method uses Markov random field to incorporate spatial information between neighboring pixels into a mixture model. This segmentation model is also learned by Expectation-Maximization algorithm using Newton-Raphson approach. The obtained results using real images data-sets are more encouraging than those obtained using similar approaches.