How to select a model in multi-way and multi-level component analysis: The CHull procedure

摘要

Principal component analysis (PCA) is a popular method for summarizing two-mode data. The last decades, several techniques have been proposed that generalize PCA to more complex types of data, like three-mode data and multi-level data. Applying these techniques in practice is often notstraightforward, as many data-analytic choices, like preprocessing, model selection, and visualization of results, become rather complicated. In this paper, we focus on the model selection problem. Specifically, we discuss recent work on the CHull procedure. This procedure looks for the component solution with the best balance between fit and complexity, where complexity is expressed via the number of free parameters. Simulation results on the petformance of the CHull procedure will be presented.