Online Program Home
My Program

Abstract Details

Activity Number: 350 - New Methods for Time Series and Longitudinal Data
Type: Contributed
Date/Time: Tuesday, July 30, 2019 : 10:30 AM to 12:20 PM
Sponsor: Section on Statistical Learning and Data Science
Abstract #305035
Title: Classification of Longitudinal Unbalanced Data: Growth Mixture Models Vs Conventional Cluster Analysis on Approximated Values at Common Time Points
Author(s): Mosammat Tanbin* and Benjamin E. Leiby and Md Jobayer Hossain
Companies: and Thomas Jefferson University and Nemours children Healthcare Systems
Keywords: Classification; longitudinal; unbalanced; Loess; interpolation; growth mixture model

Classification of unbalanced longitudinal data is a topic of growing interest in an era when naturally occurring data such as electronic medical records have shown a great utility in making personalized medical decisions. Growth Mixture Models (GMMs) can identify subgroups of homogeneous trajectories in this type of data. The method is not available in most standard software packages, requires specification of the number of classes prior to analysis, and may become computationally complex quickly. Methods available in most software packages are mainly applicable to data collected on a common set of occasions. Loess is a flexible non-parametric local regression technique that approximates a curve of best fit without assuming any distributional shape of the trajectories. It usually captures the general patterns of the curves when approximating values at fixed common time points. Conventional methods of cluster analysis can be applied on the loess approximated values considering values of each time point as a single variable. Classification accuracy using this method is similar to that using GMM. The advantage of this method over GMM is that it can handle very large dataset.

Authors who are presenting talks have a * after their name.

Back to the full JSM 2019 program