Model Insight
XGBoost Ensemble v3.2.1 · Trained on 10,542 sessions · Last retrained 7 days ago
Auto-retrain weeklyProduction
Accuracy
94.1%
+0.3%Test set n=2,847
Correct discharge timing predictions
AUC-ROC
0.972
+0.0055-fold cross-val
Discrimination power (1.0 = perfect)
F1 Score
0.921
+0.008Macro-averaged
Precision–recall harmonic mean
RMSE
2.3 wk
−0.2 wkRecovery time estimate
Lower is better
SHAP Feature Importance
Mean |SHAP| values — contribution to recovery time prediction. Red = longer recovery, Teal = faster recovery.
Positive prognosis
Risk factor
Prediction Residuals
Predicted vs (actual − predicted). Points near zero = accurate. Pattern = systematic bias.
RMSE: 2.3 wk · Bias: −0.04 wk · No systematic pattern detected
Prediction Confidence Distribution
Model confidence scores across test set (n=2,847). High confidence = reliable prediction.
Median confidence: 87%
High conf. (>90%): 39% of predictions