On Wednesday, March 13th, 2024, the seminar series hosted a talk by Prof. Paul McNicholas. Paul is the Canada research chair in computational statistics and a professor in the department of Mathematics and Statistics at McMaster University, Canada and talked about handling outliers in model based clustering. Details for the talk are below.
Title
Handling Outliers in Model-Based Clustering
Abstract
An algorithm for handling outliers in multivariate model-based clustering is introduced. The algorithm, called OCLUST, is based on the distribution of subset log-likelihoods and does not require pre-specification of the proportion of outliers. The OCLUST algorithm is demonstrated on simulated and real data before an extension to three-way data is discussed and illustrated.