|
Activity Number:
|
87
- Invited ePoster Session: a Statistical Smörgåsbord
|
|
Type:
|
Invited
|
|
Date/Time:
|
Sunday, July 29, 2018 : 8:30 PM to 10:30 PM
|
|
Sponsor:
|
Section on Statistical Learning and Data Science
|
|
Abstract #328387
|
|
|
Title:
|
Addressing Overfitting in Mixtures of Factor Analyzers
|
|
Author(s):
|
Jeffrey L Andrews*
|
|
Companies:
|
University of British Columbia Okanagan
|
|
Keywords:
|
Clustering;
Bootstrap;
EM algorithm;
Factor analyzers;
Mixture models
|
|
Abstract:
|
The expectation-maximization (EM) algorithm is a common approach for parameter estimation in the context of cluster analysis using finite mixture models. This approach suffers from the well-known issue of convergence to local maxima, but also the less obvious problem of overfitting. Mixtures of factor analyzers assume an underlying factor analysis structure and thereby perform dimensionality reduction simultaneously during the alternating expectation conditional maximization (AECM) model-fitting process. Importantly though, both convergence to local maxima and overfitting remain a concern. We address these concerns by introducing an algorithm that augments the traditional AECM with the nonparametric bootstrap. Further simulations and applications to real data lend support for the usage of this bootstrap augmented AECM-style algorithm.
|
Authors who are presenting talks have a * after their name.
