Research map / Label-efficient and robust learning / Recognition and long-tail
An Importance Weighted Feature Selection Stability Measure
Victor Hamer, Pierre Dupont
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (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 |
|---|---|---|---|
| weight feature selection stability by importance (phi_iw instead of subset phi) | stability measure · toy map of Figure 4 | stability value: 0.72 -> 0.21 | Improved |
| use phi_iw instead of phi_pears when half the importance is stable | stability measure · toy map of Figure 6 | stability value: 0.25 -> 0.5 | Improved |
| optimize phi_iw instead of phi for choosing the accuracy/stability compromise | Hybrid-RFE · singh | accuracy: 0.889 -> 0.934 | Improved |
| optimize phi_iw instead of phi_pears for the chosen compromise | Hybrid-RFE · singh | accuracy: 0.930 -> 0.934 | Improved |
| increase correlated group size c (phi, group lasso) | phi · group lasso with correlated groups | stability: phi increases when c grows | Hurt |
| increase correlated group size c (phi_pears) | phi_pears · group lasso with correlated groups | stability: phi_pears starts higher and decreases | Hurt |
| use phi_iw under growing group size c | phi_iw · group lasso with correlated groups | stability: phi_iw remains approximately constant with c | Improved |
| force selection of 5 highest-variance features with small differential shrinkage alpha=0.1 | Hybrid-RFE · singh | feature stability map | Improved |
| decrease lambda2 from 1 (lasso) toward elastic net | logistic regression · micro-array data sets | stability: all three stability measures increase when departing from lasso (lambda2 = 1) | Improved |
| increase final regularization lambda_f | logistic RFE · lsvt | phi_iw and phi_pears: phi unchanged | Improved |
| increase final regularization lambda_f strongly | logistic RFE · lsvt | accuracy: accuracy first increases then drops | Mixed |
| grow forest size from 3 trees to 3000 trees | random forest selection · all data sets | stability and accuracy: converges for T ~ 1000 | Improved |
Closest papers by meaning
- On the Stability of Feature Selection Algorithms (JMLR 2018 · similarity 0.92)
- Building a stable classifier with the inflated argmax (NeurIPS 2024 · similarity 0.85)
- Accurate Estimation of Feature Importance Faithfulness for Tree Models (AAAI 2025 · similarity 0.84)
- Estimating Conditional Mutual Information for Dynamic Feature Selection (ICLR 2024 · similarity 0.83)
- Feature Importance Metrics in the Presence of Missing Data (ICML 2025 · similarity 0.83)
- Probabilistic Stability Guarantees for Feature Attributions (NeurIPS 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.
