Activity Number:
|
281
- New Methods with Applications in Mental Health Statistics
|
Type:
|
Contributed
|
Date/Time:
|
Tuesday, August 9, 2022 : 10:30 AM to 12:20 PM
|
Sponsor:
|
Mental Health Statistics Section
|
Abstract #323029
|
|
Title:
|
Optimal Transformations of High-Dimensional Functional Data for Clustering Methods
|
Author(s):
|
Hanchao Zhang* and Thaddeus Tarpey
|
Companies:
|
NYU Grossman School of Medicine and New York University
|
Keywords:
|
Clustering;
Functional Data;
Disease Sub-typing;
Mental Health Illness;
Semi-supervised Learning
|
Abstract:
|
Clustering algorithms have been proposed for disease sub-typing for a while, especially for psychiatric illness that is not very well-defined. However, most clustering methods are completely unsupervised and perform poorly in high-dimensional settings. This talk considers a semi-supervised clustering approach that incorporates covariate information. In order to optimize the clustering, we consider pre-conditioning using linear and non-linear transformation to align and direct clustering algorithms with known covariates subgroupings. The optimization criterion for the clustering is based on the Kullback-Leibler. We propose to estimate the empirical distribution by independent component analysis (ICA) and introduce the linkage between ICA and the Naive Bayes method. This extension serves very well for psychiatric illness that has a continuous spectrum. It also has multiple applications in medical research fields that collect high-dimensional functional data. For example, related diagnostic groups may share common functional trajectories and some trajectories may be unique to a particular diagnostic group. Our approach is motivated to optimize clustering in these settings.
|
Authors who are presenting talks have a * after their name.