Evaluate Feature Selection for Leakage — evaluate_feature_selection • OmicSelector

Assesses whether a feature selection method has data leakage by checking if TRAP features were selected.

Usage

evaluate_feature_selection(selected_features, gold_standard)

Arguments

selected_features: Character vector of selected feature names
gold_standard: Output from [generate_gold_standard()]

Value

A list of class `OmicSelectorEvaluation` containing:

n_causal_selected: Number of true causal features found
n_trap_selected: Number of trap features found (should be 0)
sensitivity: Proportion of causal features recovered
precision: Proportion of selected features that are causal
leakage_score: Proportion of selected features that are traps
verdict: Human-readable assessment
pass: Logical indicating if selection passed (no traps)

Examples

if (FALSE) { # \dontrun{
gs <- generate_gold_standard(n_features = 200, n_causal = 10, n_trap = 15)

# A good selection: only causal features
good <- evaluate_feature_selection(gs$ground_truth$causal_features[1:5], gs)
print(good)  # PASS

# A leaky selection: includes trap features
bad <- evaluate_feature_selection(c("CAUSAL_01", "TRAP_01"), gs)
print(bad)  # FAIL: Data leakage detected
} # }