Figure 3.

Calibration curves of hospital admission prediction models. Lasso and gradient boosting machine showed good calibration in all models. Logistic regression was ill-calibrated for patients with the lowest and the highest deciles of predicted probability in model 4.