Close

All courses for every first-year Science student will be delivered online this fall. A limited number of students in their second, third and fourth years will return to campus for part of the semester.

Statistics Seminar - Paul McNicholas - oclust: using subset log-likelihoods to trim outliers in Gaussian mixture models

Description

Title: oclust: using subset log-likelihoods to trim outliers in Gaussian mixture models

Speaker: Paul McNicholas (McMaster University)

Abstract: Mixtures of Gaussian distributions are a popular choice in model-based clustering. Outliers can affect parameters estimation and, as such, must be accounted for. Predicting the proportion of outliers correctly is paramount as it minimizes misclassification error. It is proved that, for a finite Gaussian mixture model, the log-likelihoods of the subset models are beta-distributed. An algorithm is then proposed that predicts the proportion of outliers by measuring the adherence of a set of subset log-likelihoods to a beta reference distribution. This algorithm removes the least likely points, which are deemed outliers, until model assumptions are met.


Date/Time: Tuesday December 1, 3:30 - 4:30 

Location: Virtual 

Join Zoom Meeting
https://mcmaster.zoom.us/j/95079221401?pwd=ZHFQUTIwSUtHd1pGOGhzc01xei9TQT09

Meeting ID: 950 7922 1401
Passcode: 016314
Go Back
McMaster University - Faculty of Science | Math & Stats