Sparse Functional Data Classification via Bayesian Aggregation

Abstract: Sparse functional data frequently arise in real-world applications, posing significant challenges for accurate classification. To address this, we propose a novel classification method that integrates functional principal component analysis (FPCA) with Bayesian aggregation. Unlike traditional ensemble methods, our approach combines predicted probabilities across bootstrap replicas and refines them through Bayesian calibration using Bayesian generalized linear models (Bayesian GLMs). We evaluated the performance of the proposed method against single classifiers and conventional ensemble techniques. The simulation results demonstrate that Bayesian aggregation improves the classification accuracy over conventional methods. Finally, we validate the approach through three real-data analyses.