Using Dominance Analysis for accurate feature importance
summary
Demonstrates Azen and Budescu’s Dominance Analysis implemented in a Python library the uploader developed and claims it produces stable, comparable importance scores by computing each feature’s marginal contribution across every subset of predictors. Shows exact workflow: fit base models, enumerate combinations, calculate average incremental R² (or equivalent metric) per feature, and rank features for marginal resource allocation. Reports runtime on tens of features, behavior with correlated predictors, and examples where rankings change versus single-model coefficients.
AI-written summary · about
chapters (5)
- 0:00Introduction to talk
- 2:04Dominance analysis concept
- 5:28Calculating feature contributions
- 7:54Dominance analysis in models
- 9:13Tool and summary
related talks
43mFeature Importance and Ensemble Methods : a new perspective
40mA Practical Guide to Dimensionality Reduction Techniques
17mMaking sense of marginal effects
45mEvidence-Based Rankings
01h 05mEvaluating Model Explainability Methods
15mInterpretable Machine Learning with R
01h 33mUnderstanding and diagnosing your machine-learning models
42mInterpreting Black-Box Models with Applications to Healthcare