A1780
Title: On model-based clustering of longitudinal data with missing values
Authors: Brian Franczak - MacEwan University (Canada) [presenting]
Abstract: Missing values complicate the analysis of data with varying types. Longitudinal data consists of repeated measurements collected over time and frequently arises in many real-world applications. An expectation-maximization (EM) algorithm is developed to fit finite mixture models to longitudinal data with missing values. To account for time dependence between measurements, a modified Cholesky decomposition of the covariance matrices is utilized. These models are applied to cluster analysis, which aims to identify homogeneous subpopulations within datasets with unknown group labels. Model performance is assessed using both simulated and real datasets and evaluated using traditional metrics such as the adjusted Rand index and the Bayesian information criterion.
