Research map / Learning theory and algorithms / Kernels and regression theory
Empirical Evaluation of Resampling Procedures for Optimising SVM Hyperparameters
Jacques Wainer, Gavin Cawley
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 |
|---|---|---|---|
| use 2-fold instead of 5-fold CV for hyperparameter selection | RBF SVM · 112 UCI data sets (<10000 points) | excess loss vs kf5: -0.0012 (95% CI -0.0056, 0.0024) | No clear effect |
| use 3-fold instead of 5-fold CV for hyperparameter selection | RBF SVM · 112 UCI data sets (<10000 points) | excess loss vs kf5: -0.0031 (95% CI -0.0094, 0.0002) | No clear effect |
| use 2-fold instead of 5-fold CV to save compute | RBF SVM · 112 UCI data sets (<10000 points) | time ratio to kf5: 0.35 | Improved |
| use 3-fold instead of 5-fold CV to save compute | RBF SVM · 112 UCI data sets (<10000 points) | time ratio to kf5: 0.54 | Improved |
| use 10-fold CV or 2x repeated 5-fold or 20x bootstrap instead of 5-fold CV | RBF SVM · 112 UCI data sets (<10000 points) | excess loss vs kf5: kf10 -0.0020, 2xkf5 -0.0014, 20xboot -0.0018 | No clear effect |
| use resubstitution instead of 5-fold CV | RBF SVM · 112 UCI data sets (<10000 points) | excess loss vs kf5: 0.1354 | Hurt |
| use 20/20 hold-out instead of 5-fold CV | RBF SVM · 112 UCI data sets (<10000 points) | excess loss vs kf5: 0.0191 | Hurt |
| use 10/10 hold-out instead of 5-fold CV | RBF SVM · 112 UCI data sets (<10000 points) | excess loss vs kf5: 0.0255 | Hurt |
| use 80/20 hold-out instead of 5-fold CV | RBF SVM · 112 UCI data sets (<10000 points) | excess loss vs kf5: 0.0053 (95% CI -0.0003, 0.0115) | No clear effect |
| use kf2 instead of kf5 on the 9 largest data sets | RBF SVM · 9 largest UCI data sets (>10000 points) | excess loss vs kf5: 0.0073 on chess-krvk; -0.0004 on pendigits | Mixed |
| use kf2 or kf3 instead of kf5 with polynomial and linear kernels | SVM · 112 UCI data sets (<10000 points) (polynomial and linear kernels) | excess loss vs kf5: kf2 0.006 polynomial, 0.002 linear | No clear effect |
| repeat the bootstrap 5 times instead of 20 times | RBF SVM · 112 UCI data sets (<10000 points) | time ratio to kf5: 4.43 -> 1.24 | Improved |
Closest papers by meaning
- Improving Efficiency of SVM k-Fold Cross-Validation by Alpha Seeding (AAAI 2017 · similarity 0.85)
- Kernel Functional Optimisation (NeurIPS 2021 · similarity 0.85)
- Reshuffling Resampling Splits Can Improve Generalization of Hyperparameter Optimization (NeurIPS 2024 · similarity 0.83)
- Restoring balance: principled under/oversampling of data for optimal classification (ICML 2024 · similarity 0.83)
- Taming Hyperparameter Sensitivity in Data Attribution: Practical Selection Without Costly Retraining (NeurIPS 2025 · similarity 0.82)
- Reproduce, Replicate, Reevaluate. The Long but Safe Way to Extend Machine Learning Methods (AAAI 2024 · 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.
