Colling 2026

Publication in Journal of Open Source Software, 11(118), 9539.

Abstract

corrselect is an R package for reducing multicollinearity and redundancy in predictor sets. It provides two complementary approaches: high-level pruning functions that return a single optimal subset, and exhaustive enumeration of all maximal admissible subsets. The package accommodates both numeric and mixed-type data, permits forced inclusion of essential predictors, and works with standard R modeling workflows including mixed-effects models. Version 3.0 introduces corrPrune() for association-based pruning and modelPrune() for VIF-based model pruning, while maintaining the original exhaustive enumeration functions (corrSelect(), assocSelect(), MatSelect()). A fast C++ greedy algorithm enables efficient pruning for large predictor sets (p > 100), while exact graph-theoretic algorithms guarantee complete enumeration when exhaustive search is feasible.