Research map / Learning theory and algorithms / Prediction and decision losses
Estimating the Replication Probability of Significant Classification Benchmark Experiments
Daniel Berrar
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 |
|---|---|---|---|
| bootstrap the standard deviation of Zw instead of fixing it to 1 (ablation) | Wilcoxon replication-probability estimator · Study 1 (20 datasets) | mean absolute deviance: 0.07318 -> 0.009792 | Improved |
| use Wilcoxon model instead of binomial model | replication probability estimate · Study 1 group 3 | empirical replication probability: 0.5246 -> 0.778 | Mixed |
| lower the p-value from 0.05 toward 0.00274 | significance threshold · Study 1 | replication probability: 0.95696 | Improved |
| use the Bayesian model instead of the binomial model | replication probability estimate · Study 1 (20 datasets) | bias vs empirical | Mixed |
| increase oracle label fraction from q=3 to q=7 | SVMo vs SVM · Study 3 | empirical replication probability: 0.495 -> 0.96597 | Improved |
| increase the number of features from d=11 to d=24 | SVM vs RF · Study 4 | empirical replication probability: 0.4484 -> 0.95896 | Improved |
| use estimator at small p-values near 0.05 (d=11) | replication probability estimator · Study 4 | estimate vs empirical: 0.5235 vs 0.4484 | Hurt |
| increase datasets from 20 to 44 | Wilcoxon estimator · Studies 1 and 2 | mean absolute deviance: 0.009792 -> 0.0162 | No clear effect |
Closest papers by meaning
- Reproduce, Replicate, Reevaluate. The Long but Safe Way to Extend Machine Learning Methods (AAAI 2024 · similarity 0.85)
- Boosting Test Performance with Importance Sampling--a Subpopulation Perspective (AAAI 2025 · similarity 0.82)
- Calibration without labels in multiple testing (NeurIPS 2026 · similarity 0.82)
- Quantifying Uncertainty in Error Consistency: Towards Reliable Behavioral Comparison of Classifiers (NeurIPS 2025 · similarity 0.82)
- Task-Agnostic Machine-Learning-Assisted Inference (NeurIPS 2024 · similarity 0.81)
- Crossing the Validation crisis: Cross-validation reduces benchmarking variance surprisingly well (NeurIPS 2026 · similarity 0.81)
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.
