Probability calibration and robustness
Calibration and sensitivity
Cross-fitted calibration comparisons and prespecified sensitivity estimands.
Evidence scope
Scope, workflow status, and summary measures
- Selected XGBoost evaluation calibrator
- logistic Cross Fitted Calibration
- Selected LightGBM evaluation calibrator
- logistic Cross Fitted Calibration
- Selected CatBoost evaluation calibrator
- none Cross Fitted Calibration
- Calibration selection rule
- Select a method only when the animal-clustered 95% CI for mean cumulative Brier improvement excludes zero and pooled log loss does not worsen. Cross Fitted Calibration
- Rows per framework
- 396 rows Prediction Sensitivity Analysis
- Sensitivity scenarios
- 2 scenarios Prediction Sensitivity Analysis
- Sensitivity estimands
- 3 estimands Prediction Sensitivity Analysis
Aggregate estimate plot
Calibration log loss
Cross Fitted CalibrationUncertainty: Point estimates are shown because this table does not supply matching lower and upper interval columns.
View plotted values in tabular form
| Estimate | Marker | Log Loss |
|---|---|---|
| catboost · isotonic | ▲ | 1.730 |
| catboost · logistic | ✦ | 1.030 |
| catboost | ⬟ | 0.968 |
| lgbm · isotonic | ■ | 1.528 |
| lgbm · logistic | ◆ | 0.793 |
| lgbm | ✦ | 0.816 |
| xgb · isotonic | ● | 1.448 |
| xgb · logistic | ▲ | 0.815 |
| xgb | ◆ | 0.844 |
Aggregate estimate plot
Sensitivity confidence intervals
Prediction Sensitivity AnalysisUncertainty: Horizontal intervals use Ci Lower and Ci Upper from the published aggregate table.
View plotted values in tabular form
| Estimate | Marker | Estimate | Lower interval | Upper interval |
|---|---|---|---|---|
| xgb · uncalibrated | ● | 0.852 | 0.802 | 0.898 |
| xgb · uncalibrated | ● | 0.817 | 0.782 | 0.854 |
| xgb · uncalibrated | ● | 0.835 | 0.801 | 0.869 |
| xgb · uncalibrated | ● | 0.621 | 0.570 | 0.674 |
| xgb · uncalibrated | ● | 0.580 | 0.524 | 0.637 |
| xgb · uncalibrated | ● | 0.844 | 0.778 | 0.908 |
| xgb · uncalibrated | ● | 0.467 | 0.389 | 0.545 |
| xgb · uncalibrated | ● | 0.154 | 0.139 | 0.170 |
| xgb · uncalibrated | ● | 0.852 | 0.794 | 0.900 |
| xgb · uncalibrated | ● | 0.817 | 0.776 | 0.857 |
| xgb · uncalibrated | ● | 0.835 | 0.795 | 0.869 |
| xgb · uncalibrated | ● | 0.621 | 0.563 | 0.682 |
| xgb · uncalibrated | ● | 0.580 | 0.520 | 0.638 |
| xgb · uncalibrated | ● | 0.844 | 0.766 | 0.915 |
| xgb · uncalibrated | ● | 0.467 | 0.383 | 0.553 |
| xgb · uncalibrated | ● | 0.154 | 0.137 | 0.172 |
| lgbm · uncalibrated | ■ | 0.859 | 0.810 | 0.902 |
| lgbm · uncalibrated | ■ | 0.830 | 0.793 | 0.869 |
| lgbm · uncalibrated | ■ | 0.845 | 0.811 | 0.878 |
| lgbm · uncalibrated | ■ | 0.636 | 0.586 | 0.687 |
| lgbm · uncalibrated | ■ | 0.596 | 0.541 | 0.649 |
| lgbm · uncalibrated | ■ | 0.816 | 0.755 | 0.875 |
| lgbm · uncalibrated | ■ | 0.457 | 0.387 | 0.529 |
| lgbm · uncalibrated | ■ | 0.149 | 0.134 | 0.163 |
Pooled calibration metrics
Summary method: approved_rows
- Indexed rows
- 9
- Eligible rows
- 9
Showing 1–9 of 9 matching rows
Filters combine with AND. Empty and unreported values sort last.
| catboost | isotonic | 396 | 115 | 0.869 | 0.845 | 0.857 | 0.652 |
|---|---|---|---|---|---|---|---|
| catboost | logistic | 396 | 115 | 0.873 | 0.848 | 0.861 | 0.634 |
| catboost | — | 396 | 115 | 0.871 | 0.852 | 0.861 | 0.644 |
| lgbm | isotonic | 396 | 115 | 0.858 | 0.837 | 0.848 | 0.669 |
| lgbm | logistic | 396 | 115 | 0.864 | 0.850 | 0.857 | 0.662 |
| lgbm | — | 396 | 115 | 0.859 | 0.830 | 0.845 | 0.636 |
| xgb | isotonic | 396 | 115 | 0.851 | 0.837 | 0.844 | 0.634 |
| xgb | logistic | 396 | 115 | 0.861 | 0.845 | 0.853 | 0.641 |
| xgb | — | 396 | 115 | 0.852 | 0.817 | 0.835 | 0.621 |
Paired calibration-method comparisons
Summary method: approved_rows
- Indexed rows
- 12
- Eligible rows
- 12
Showing 1–12 of 12 matching rows
Filters combine with AND. Empty and unreported values sort last.
| catboost | logistic | log_loss | uncalibrated_minus_calibrated | -0.061 | -0.324 | 0.187 | 0.378 |
|---|---|---|---|---|---|---|---|
| catboost | logistic | mean_cumulative_brier | uncalibrated_minus_calibrated | 0.001 | -0.002 | 0.004 | 0.812 |
| catboost | isotonic | log_loss | uncalibrated_minus_calibrated | -0.762 | -1.325 | -0.265 | 0.002 |
| catboost | isotonic | mean_cumulative_brier | uncalibrated_minus_calibrated | -2.26e-4 | -0.005 | 0.005 | 0.463 |
| lgbm | logistic | log_loss | uncalibrated_minus_calibrated | 0.023 | -0.006 | 0.052 | 0.935 |
| lgbm | logistic | mean_cumulative_brier | uncalibrated_minus_calibrated | 0.007 | 6.52e-4 | 0.013 | 0.985 |
| lgbm | isotonic | log_loss | uncalibrated_minus_calibrated | -0.713 | -1.356 | -0.213 | 6.00e-4 |
| lgbm | isotonic | mean_cumulative_brier | uncalibrated_minus_calibrated | 0.003 | -0.004 | 0.010 | 0.769 |
| xgb | logistic | log_loss | uncalibrated_minus_calibrated | 0.030 | -0.002 | 0.063 | 0.970 |
| xgb | logistic | mean_cumulative_brier | uncalibrated_minus_calibrated | 0.009 | 0.003 | 0.017 | 0.999 |
| xgb | isotonic | log_loss | uncalibrated_minus_calibrated | -0.604 | -1.060 | -0.199 | 2.00e-4 |
| xgb | isotonic | mean_cumulative_brier | uncalibrated_minus_calibrated | 0.007 | -1.59e-4 | 0.014 | 0.971 |
Sensitivity estimands
Summary method: approved_rows
- Indexed rows
- 18
- Eligible rows
- 18
Showing 1–18 of 18 matching rows
Filters combine with AND. Empty and unreported values sort last.
| selected_calibration | xgb | row_weighted_pooled | 0.861 | 0.845 | 0.853 | 0.641 | 0.609 |
|---|---|---|---|---|---|---|---|
| selected_calibration | xgb | equal_animal_weighted | 0.847 | 0.869 | 0.858 | 0.730 | 0.623 |
| selected_calibration | xgb | equal_outer_fold_weighted | 0.865 | 0.854 | 0.860 | 0.641 | 0.609 |
| selected_calibration | lgbm | row_weighted_pooled | 0.864 | 0.850 | 0.857 | 0.662 | 0.628 |
| selected_calibration | lgbm | equal_animal_weighted | 0.847 | 0.871 | 0.859 | 0.757 | 0.644 |
| selected_calibration | lgbm | equal_outer_fold_weighted | 0.869 | 0.861 | 0.865 | 0.662 | 0.628 |
| selected_calibration | catboost | row_weighted_pooled | 0.871 | 0.852 | 0.861 | 0.644 | 0.606 |
| selected_calibration | catboost | equal_animal_weighted | 0.851 | 0.871 | 0.861 | 0.740 | 0.612 |
| selected_calibration | catboost | equal_outer_fold_weighted | 0.876 | 0.860 | 0.868 | 0.644 | 0.605 |
| uncalibrated | xgb | row_weighted_pooled | 0.852 | 0.817 | 0.835 | 0.621 | 0.580 |
| uncalibrated | xgb | equal_animal_weighted | 0.834 | 0.837 | 0.835 | 0.719 | 0.585 |
| uncalibrated | xgb | equal_outer_fold_weighted | 0.865 | 0.854 | 0.860 | 0.621 | 0.581 |
| uncalibrated | lgbm | row_weighted_pooled | 0.859 | 0.830 | 0.845 | 0.636 | 0.596 |
| uncalibrated | lgbm | equal_animal_weighted | 0.841 | 0.843 | 0.842 | 0.734 | 0.597 |
| uncalibrated | lgbm | equal_outer_fold_weighted | 0.869 | 0.861 | 0.865 | 0.637 | 0.597 |
| uncalibrated | catboost | row_weighted_pooled | 0.871 | 0.852 | 0.861 | 0.644 | 0.606 |
| uncalibrated | catboost | equal_animal_weighted | 0.851 | 0.871 | 0.861 | 0.740 | 0.612 |
| uncalibrated | catboost | equal_outer_fold_weighted | 0.876 | 0.860 | 0.868 | 0.644 | 0.605 |
Sensitivity confidence intervals
Summary method: approved_rows
- Indexed rows
- 96
- Eligible rows
- 96
Showing 1–25 of 96 matching rows
Filters combine with AND. Empty and unreported values sort last.
| selected_calibration | xgb | early_auc | 0.861 | 0.817 | 0.901 | 0.021 | 5,000 |
|---|---|---|---|---|---|---|---|
| selected_calibration | xgb | late_auc | 0.845 | 0.811 | 0.879 | 0.017 | 5,000 |
| selected_calibration | xgb | mean_auc | 0.853 | 0.822 | 0.882 | 0.015 | 5,000 |
| selected_calibration | xgb | accuracy | 0.641 | 0.591 | 0.692 | 0.026 | 5,000 |
| selected_calibration | xgb | balanced_accuracy | 0.609 | 0.555 | 0.664 | 0.028 | 5,000 |
| selected_calibration | xgb | log_loss | 0.815 | 0.739 | 0.891 | 0.039 | 5,000 |
| selected_calibration | xgb | ordinal_mae | 0.437 | 0.368 | 0.508 | 0.036 | 5,000 |
| selected_calibration | xgb | mean_cumulative_brier | 0.145 | 0.129 | 0.161 | 0.008 | 5,000 |
| selected_calibration | xgb | early_auc | 0.861 | 0.809 | 0.903 | 0.025 | 5,000 |
| selected_calibration | xgb | late_auc | 0.845 | 0.807 | 0.882 | 0.019 | 5,000 |
| selected_calibration | xgb | mean_auc | 0.853 | 0.817 | 0.883 | 0.017 | 5,000 |
| selected_calibration | xgb | accuracy | 0.641 | 0.585 | 0.700 | 0.029 | 5,000 |
| selected_calibration | xgb | balanced_accuracy | 0.609 | 0.548 | 0.668 | 0.031 | 5,000 |
| selected_calibration | xgb | log_loss | 0.815 | 0.727 | 0.897 | 0.044 | 5,000 |
| selected_calibration | xgb | ordinal_mae | 0.437 | 0.359 | 0.514 | 0.040 | 5,000 |
| selected_calibration | xgb | mean_cumulative_brier | 0.145 | 0.127 | 0.162 | 0.009 | 5,000 |
| selected_calibration | lgbm | early_auc | 0.864 | 0.817 | 0.906 | 0.022 | 5,000 |
| selected_calibration | lgbm | late_auc | 0.850 | 0.820 | 0.881 | 0.016 | 5,000 |
| selected_calibration | lgbm | mean_auc | 0.857 | 0.827 | 0.886 | 0.015 | 5,000 |
| selected_calibration | lgbm | accuracy | 0.662 | 0.613 | 0.710 | 0.025 | 5,000 |
| selected_calibration | lgbm | balanced_accuracy | 0.628 | 0.575 | 0.680 | 0.027 | 5,000 |
| selected_calibration | lgbm | log_loss | 0.793 | 0.720 | 0.869 | 0.038 | 5,000 |
| selected_calibration | lgbm | ordinal_mae | 0.409 | 0.341 | 0.480 | 0.035 | 5,000 |
| selected_calibration | lgbm | mean_cumulative_brier | 0.142 | 0.126 | 0.159 | 0.008 | 5,000 |
| selected_calibration | lgbm | early_auc | 0.864 | 0.809 | 0.908 | 0.025 | 5,000 |
Uncalibrated row-weighted pooled framework ranking
Summary method: approved_rows
- Indexed rows
- 3
- Eligible rows
- 3
Showing 1–3 of 3 matching rows
Filters combine with AND. Empty and unreported values sort last.
| 1 | catboost | 0.861 | 0.606 | 0.407 |
|---|---|---|---|---|
| 2 | lgbm | 0.845 | 0.596 | 0.457 |
| 3 | xgb | 0.835 | 0.580 | 0.467 |