featR

A unified R package of feature-selection methods: statistical filters, regularization, model-based wrappers, and dimensionality reduction behind a consistent set of fs_*() functions.

Heavy modeling engines (brms, caret, glmnet, randomForest, ...) are optional Suggests — each function checks for what it needs and tells you what to install.

Installation

# From a local checkout:
# install.packages("devtools")
devtools::install()

# Or build/check from the shell:
# R CMD build . && R CMD check --as-cran featR_0.1.0.tar.gz

Functions

1. Statistical & Filter Methods

Function	What it does	Notes
fs_chi	Chi-square tests of association between categorical features and a categorical target, with p-value adjustment, Yates/Monte-Carlo handling.	Returns a results table plus significant_features.
fs_correlation	Computes a correlation matrix (Pearson/Spearman/Kendall/point-biserial/polychoric) and flags variable pairs above a threshold.	Returns both members of each high-correlation pair (the redundant set) — prune accordingly.
fs_infogain	Information gain of each feature w.r.t. a target; supports numeric (binned), categorical, and date features.	The target is discretized once so scores are comparable across features.
fs_supervised	Threshold filter scoring features against a target: absolute Pearson correlation (numeric target) or ANOVA F (factor target).	Flexible output shapes (matrix, data.table, mask, indices, names, list).
fs_unsupervised	Target-free threshold filter: variance, MAD, IQR, range, missing proportion, or distinct-value count.	Variance filtering lives here (method = "variance").

2. Regularization Methods

Function	What it does	Notes
fs_elastic	Elastic-net over an alpha/lambda grid via caret + glmnet, optional PCA preprocessing.	Reports best alpha/lambda and coefficients.
fs_lasso	Cross-validated LASSO (glmnet) returning coefficient-based importance at lambda.min.	Importance is on the original predictor scale — see docs.

3. Model-Based & Wrapper Methods

Function	What it does	Notes
fs_bayes	Bayesian model comparison over predictor combinations (brms), ranked by LOO.	Expensive; MAE/RMSE reported are in-sample.
fs_boruta	Boruta all-relevant selection with optional correlation-based pruning.
fs_randomforest	Random-forest train/evaluate pipeline with permutation importance and preprocessing options.
fs_recursivefeature	caret RFE on a training split, evaluated on a held-out test split; optional final model on training data.
fs_stepwise	Stepwise linear regression via MASS::stepAIC (forward/backward/both).	Post-selection p-values are not valid for inference — see docs.
fs_svm	SVM (caret) train/evaluate pipeline with optional feature selection and class-imbalance handling.	Feature selection uses random-forest RFE, not SVM-RFE. Classification returns a confusion matrix; regression returns RMSE/R2/MAE.
fs_mars	MARS (earth) train/evaluate pipeline with tuning and optional ROC/PR AUC.

4. Dimensionality Reduction

Function	What it does	Notes
fs_pca	PCA (prcomp, or bigstatsr for large data) with loadings, scores, variance explained, optional plot.
fs_svd	Exact or approximate (RSpectra) truncated SVD.	Request n_singular_values < min(dim(x)) to enable the approximate solver.

Conventions

All functions validate inputs up front, are sequential by default (parallelism is opt-in and capped), never seed the RNG unless you pass seed = ... (and then restore its state), and signal progress with suppressible message()s.

License

MIT (c) Justin Chase. See LICENSE.md.