Research map / Learning theory and algorithms / Kernels and regression theory
Randomization Can Reduce Both Bias and Variance: A Case Study in Random Forests
Brian Liu, Rahul Mazumder
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 |
|---|---|---|---|
| add split feature subsetting (mtry 0.5 instead of 1.0) (ablation) | bagging ensemble · Hidden2D (SNR 6) | test MSE: 0.047 -> 0.029 | Improved |
| add split feature subsetting (mtry 0.5) (ablation) | bagging ensemble · Hidden2D (SNR 6) | training MSE: 0.007 -> 0.005 | Improved |
| add split feature subsetting (mtry 0.5) (ablation) | bagging ensemble · Hidden2D (SNR 6) | squared bias: 0.00525 -> 0.00235 | Improved |
| add split feature subsetting (mtry 0.5) (ablation) | bagging ensemble · Hidden2D (SNR 6) | variance: 0.00838 -> 0.00493 | Improved |
| add split feature subsetting (mtry 0.5) (ablation) | bagging ensemble · HiddenSine (SNR 6) | squared bias: 0.0876 -> 0.06786 | Improved |
| add split feature subsetting (mtry 0.333) (ablation) | bagging ensemble · HiddenCube (SNR 6) | squared bias: 0.1300 -> 0.1200 | Improved |
| add split feature subsetting (mtry 0.333) (ablation) | bagging ensemble · HiddenSquare (SNR 6) | variance: 0.002439 -> 0.000871 | Improved |
| trim bagging trees (maxnodes tuned to match effective DoF of random forest) (ablation) | bagging ensemble · MARS and MARSadd simulations | test error | No clear effect |
| add augmented bagging (noise features) to match effective DoF (ablation) | bagging ensemble · MARS and MARSadd simulations | test error | No clear effect |
| add split feature subsetting (mtry 0.333) (ablation) | bagging ensemble · Hidden2D, SNR below 1 | training error | Hurt |
| add split feature subsetting (ablation) | bagging ensemble · Hidden2D, increasing SNR | test MSE decrease | Improved |
| increase dataset size n from 1000 to 20000 with random forest instead of bagging (ablation) | random forest vs bagging · Polynomial Hidden2D (SNR 6) | squared bias: 1.5128 -> 1.3861 (n=1000), 0.9383 -> 0.4605 (n=20000) | Improved |
Closest papers by meaning
- Randomization as Regularization: A Degrees of Freedom Explanation for Random Forest Success (JMLR 2020 · similarity 0.91)
- Exogenous Randomness Empowering Random Forests (JMLR 2026 · similarity 0.89)
- Getting Better from Worse: Augmented Bagging and A Cautionary Tale of Variable Importance (JMLR 2022 · similarity 0.89)
- Understanding Randomization in Greedy Model Search (NeurIPS 2026 · similarity 0.88)
- Generalized Random Forests Using Fixed-Point Trees (ICML 2025 · similarity 0.85)
- DynFrs: An Efficient Framework for Machine Unlearning in Random Forest (ICLR 2025 · similarity 0.85)
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.
