Y-Randomization Validation Results for Predictive Models

Results from 100 Y-randomization iterations per model-predictor-target combination to assess predictive model validity. The dataset includes metrics like R2_fit_random, Q2_LOOCV_random, and p_value_random for models fit on randomized target variables. Authored by Mythili V and shared on figshare in April 2026.

Use Cases

• Compare original model performance against R2_fit_random and Q2_LOOCV_random scores from 100 randomized iterations to detect overfitting.
• Analyze the distribution of p_value_random across multiple Model and Target combinations to assess statistical significance of original predictions.
• Benchmark the robustness of different Model Families by comparing their RMSE_LOOCV_random and MAE_LOOCV_random metrics under randomization.
• Use the Iteration, Predictor, and Target fields to study the stability of chance correlations across different experimental setups.

Strengths

• Contains 100 randomization iterations for each model-predictor-target combination, providing a substantial basis for statistical analysis.
• Includes multiple validation metrics per iteration: R2_fit_random, Q2_LOOCV_random, RMSE_LOOCV_random, MAE_LOOCV_random, r_corr_random, and p_value_random.

Limitations

• The total number of rows (observations) is unknown, making it difficult to assess the overall dataset scale.
• As a 4.6 MB file, the dataset is relatively small and may represent results from a specific, limited set of models or domains.
• Lacks the original, non-randomized model performance data for direct within-dataset comparison.

Provenance

Source

figshare, author Mythili V.

Collection Method

Generated from Y-randomization (response permutation) experiments on predictive models.

Freshness

Last updated in April 2026.