Clustering incomplete longitudinal data
Mateen R. Shaikh · The Atrium (University of Guelph) · 2009
A model-based approach to the clustering of incomplete longitudinal data is presented. Clustering attempts to find groups within data when nothing about group memberships is known while incomplete longitudinal data arises in many situations and complicates statistical analyses. Although one common practice is to delete incomplete observations, this is undesirable for many reasons. Utilizing incomplete observations is difficult and in the context of clustering poses problems in existing methods. The framework of model-based clustering and in particular, an existing model that utilizes the modified Cholesky decomposition is exploited. A variant of the EM algorithm is used for parameter estimation and the BIC is used for model selection. This approach is illustrated on real data, including real gene expression time course data.