|
Activity Number:
|
144
- Digital Phenotyping
|
|
Type:
|
Invited
|
|
Date/Time:
|
Monday, July 29, 2019 : 10:30 AM to 12:20 PM
|
|
Sponsor:
|
Mental Health Statistics Section
|
|
Abstract #300215
|
Presentation
|
|
Title:
|
Supervised Kernel PCA for Longitudinal Data in Mental Health
|
|
Author(s):
|
Gregory Ryslik and Patrick Staples* and Min Ouyang and Paul Dagum
|
|
Companies:
|
Mindstrong Health and Mindstrong Health and Mindstrong Health
|
|
Keywords:
|
Supervised Kernel PCA;
longitudinal models,;
digital biomarkers;
dimension reduction;
digital biomarkers;
kernel methods
|
|
Abstract:
|
With the development of the smartphone, the field of digital phenotyping now allows for passive ecological continuous monitoring of patients suffering from mental disorders. The data collected from such devices creates a high-dimensional multivariate time series per patient. In order to model the relationship between such data and an outcome of interest it is often necessary to reduce the data dimensionality with respect to a specific target. We present an extension to supervised kernel PCA for longitudinal data, which enables the reduction of phenotype data into separate within- and between-subject components. In our applications, this technique combined with mixed modeling yields superior predictive accuracy compared to the model it extends.
|
Authors who are presenting talks have a * after their name.
