Skip to main content



Master Thesis Defense - February 28, 2019: Finite Bivariate and Multivariate Beta Mixture Models Learning and Applications

February 14, 2019


Narges Manouchehri

Thursday, February 28, 2019 at 2:00 p.m.
Room EV001.162

You are invited to attend the following M.A.Sc. (Quality Systems Engineering) thesis examination.

Examining Committee

Dr. W. Lucia, Chair
Dr. N. Bouguila, Supervisor
Dr. F. Mafakheri, CIISE Examiner
Dr. J. Lee, External Examiner (BCEE)



Finite mixture models have been revealed to provide flexibility for data clustering. They have demonstrated high competence and potential to capture hidden structure in data. Modern technological progresses, growing volumes and varieties of generated data, revolutionized computers and other related factors are contributing to produce large scale data. This fact enhances the significance of finding reliable and adaptable models which can analyze bigger, more complex data to identify latent patterns, deliver faster and more accurate results and make decisions with minimal human interaction. Adopting the finest and most accurate distribution that appropriately represents the mixture components is critical. The most widely adopted generative model has been the Gaussian mixture. In numerous real-world applications, however, when the nature and structure of data are non-Gaussian, this modeling fails. One of the other crucial issues when using mixtures is determination of the model complexity or number of mixture components. Minimum message length (MML) is one of the main techniques in frequentist frameworks to tackle this challenging issue. In this work, we have designed and implemented a finite mixture model, using the bivariate and multivariate Beta distributions for cluster analysis and demonstrated its flexibility in describing the intrinsic characteristics of the observed data. In addition, we have applied our estimation and model selection algorithms to synthetic and real datasets. Most importantly, we considered interesting applications such as in image segmentation, software modules defect prediction, spam detection and occupancy estimation in smart buildings.

Back to top Back to top

© Concordia University