corrselect: Correlation-Based and Model-Based Predictor Pruning

Provides tools for reducing multicollinearity in predictor sets through association-based and model-based approaches. The package offers both fast greedy algorithms for quick pruning and exact graph-theoretic algorithms for exhaustive subset enumeration.

Association-Based Pruning

These functions identify variable subsets where all pairwise correlations or associations remain below a user-defined threshold:

corrPrune

Fast greedy pruning for numeric data

corrSelect

Exhaustive enumeration for numeric data frames

assocSelect

Exhaustive enumeration for mixed-type data (numeric, factor, ordered)

MatSelect

Direct interface using a pre-computed correlation matrix

Model-Based Pruning

These functions use variance inflation factors (VIF) to iteratively remove collinear predictors from regression models:

modelPrune

VIF-based pruning for lm, glm, lme4, and glmmTMB models

Algorithms

The exact enumeration functions (corrSelect, assocSelect, MatSelect) use two graph-theoretic algorithms:

Eppstein-Loffler-Strash (ELS)

Recommended when using force_in constraints

Bron-Kerbosch

Default algorithm, with optional pivoting for performance

Helpers

corrSubset

Extract specific subsets from results

CorrCombo-class

S4 class holding enumeration results

See also

Vignettes: vignette("quickstart", package = "corrselect"), vignette("advanced", package = "corrselect")

Author

Maintainer: Gilles Colling gilles.colling051@gmail.com