Research map / Learning theory and algorithms / Prediction and decision losses
A Debiased MDI Feature Importance Measure for Random Forests
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| Evaluate feature contributions on out-of-bag samples rather than the tree's training samples | MDI feature importance · Simulated classification, deep trees | AUC: 0.12 -> 0.76 | Helped |
| Evaluate feature contributions on out-of-bag samples rather than the tree's training samples | MDI feature importance · Simulated regression, deep trees | AUC: 0.09 -> 0.52 | Helped |
| Evaluate feature contributions on out-of-bag samples rather than the tree's training samples | MDI feature importance · ChIP classification, deep trees | AUC: 0.60 -> 0.87 | Helped |
| Evaluate feature contributions on out-of-bag samples rather than the tree's training samples | MDI feature importance · ChIP regression, deep trees | AUC: 0.71 -> 0.98 | Helped |
| Increase minimum leaf size from 1 to 100 (ablation) | MDI-oob · Simulated classification | AUC: 0.76 -> 0.75 | Hurt |
| Increase minimum leaf size from 1 to 100 (ablation) | MDI-oob · Simulated regression | AUC: 0.52 -> 0.58 | Helped |
| Increase minimum leaf size from 1 to 100 (ablation) | MDI-oob · ChIP classification | AUC: 0.87 -> 0.94 | Helped |
| Increase minimum leaf size from 1 to 100 (ablation) | MDI-oob · ChIP regression | AUC: 0.98 -> 0.98 | No clear effect |
| Use UFI instead of MDI-oob | Random Forest feature selection · Simulated regression, deep trees | AUC: 0.52 -> 0.54 | Helped |
| Use UFI instead of MDI-oob | Random Forest feature selection · ChIP classification, deep trees | AUC: 0.87 -> 0.88 | Helped |
| Use UFI instead of MDI-oob | Random Forest feature selection · ChIP regression, deep trees | AUC: 0.98 -> 0.99 | Helped |
Closest papers by meaning
- From global to local MDI variable importances for random forests and when they are Shapley values (NeurIPS 2021 · similarity 0.89)
- Variable Importance Using Decision Trees (NeurIPS 2017 · similarity 0.86)
- Accurate Estimation of Feature Importance Faithfulness for Tree Models (AAAI 2025 · similarity 0.85)
- Conditional Feature Importance with Generative Modeling Using Adversarial Random Forests (AAAI 2025 · similarity 0.85)
- Multistage Defer Trees for Hybrid Interpretability: If at First You Can't Succeed, Tree Again (NeurIPS 2026 · similarity 0.82)
- Feature Importance Metrics in the Presence of Missing Data (ICML 2025 · similarity 0.82)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.