Compute Permutation Feature Importance — feature

Computes model-agnostic feature importance by permuting features and measuring the drop in performance.

Usage

feature_importance(
  explainer,
  loss_function = c("auc_loss", "accuracy_loss", "brier"),
  n_permutations = 10,
  parallel = FALSE
)

Arguments

explainer: An OmicExplainer object
loss_function: Loss function to use (default: "1 - AUC")
n_permutations: Number of permutations per feature (default: 10)
parallel: Logical, use parallel computation (default: FALSE)

Value

A data frame with feature importance scores

Details

Permutation importance measures how much model performance degrades when a feature's values are randomly shuffled. Higher values indicate more important features.

For correlated features, importance may be shared or masked. Use `feature_importance_groups()` for correlated feature handling.

Examples

if (FALSE) { # \dontrun{
importance <- feature_importance(explainer, n_permutations = 20)
head(importance[order(-importance$importance), ])
} # }