feature_importance

Sampling, Validation and ML Diagnostics

Feature ranking methods: MDI, MDA, and single-feature importance with PCA diagnostics.

Why This Module Exists

Improves model interpretability and helps remove unstable or redundant features.

Key Public APIs

mean_decrease_impurity
mean_decrease_accuracy
single_feature_importance
feature_pca_analysis

Core Math

MDI

\[I_j=\sum_{t\in T_j} p(t)\Delta i(t)\]

MDA

\[I_j=Score(X)-Score(X_{perm(j)})\]

Code Examples

Run MDA with classifier

use openquant::feature_importance::mean_decrease_accuracy;

// Plug in your classifier implementing SimpleClassifier
let importance = mean_decrease_accuracy(&clf, &x, &y, 5)?;

Implementation Notes

Cross-validated MDA is preferred when leakage risk is high.
Compare ranking stability across folds/time windows.