Variable selection for clustering and classification of data with missing values

Abstract

This poster presentation embarks on a comprehensive exploration of explicit variable selection procedures in model-based classification, where classification aims to assign labels to unlabelled observations. Delving into existing methodologies, we will dissect the intricacies of variable selection, setting the stage for an extensive examination of an approach aimed at minimizing within-group variance while maximizing between-group variance, known as Variable Selection for Clustering and Classification (VSCC). With a focus on enhancing classification accuracy and interpretability, we will unveil the details of VSCC, elucidating its significance in model-based classification frameworks. Furthermore, we will investigate how this approach performs when applied to simulated and real data sets with missing values. Through meticulous evaluation and analysis, we will scrutinize the performance and robustness of the variable selection approach in handling the challenges posed by incomplete data. Our findings will be synthesized into a comprehensive discussion, shedding light on the implications of the results and offering valuable insights for future research directions and refinements in variable selection methodologies within model-based classification.