Fast and Scalable Cellwise-Robust Ensembles for High-Dimensional Data

Variable selection and ensemble methods are central to high-dimensional modelling, enabling the identification of relevant predictors and the construction of stable predictive signals through aggregation across multiple models. However, in practice, high-dimensional data are often affected by cellwise contamination, in which individual cells of the data matrix deviate from the underlying multivariate structure without necessarily making the corresponding observation outlying. This type of contamination can easily propagate throughout many observations, compromising variable selection procedures and ensemble methods, including robust methods designed for contamination affecting entire observations (casewise contamination). To address this limitation, we propose the Fast and Scalable Cellwise-Robust Ensemble (FSCRE) algorithm. FSCRE dynamically partitions predictors into disjoint sub-models using a competitive proposer-arbiter architecture operating in a robust correlation framework. Through extensive simulations and a bioinformatics application, we demonstrate FSCRE's competitive performance in variable selection precision, recall, and predictive accuracy across various contamination scenarios, all while maintaining high computational efficiency in high-dimensional settings. This work provides a unified framework connecting cellwise-robust estimation with high-performance ensemble learning, with an implementation available on CRAN.