Figure 1
Models predictive performance and calibration. Fastai (F) tabular model performance is depicted as the purple line; fastai tabular-textual model is the cyan line; the XGBoost (XGB) model is the blue line; the TensorFlow (TF) tabular model is depicted as the green line; the TensorFlow model is the red line. (a) Receiver operating characteristic (ROC) curves of models for 30-day mortality outcome. (c) ROC curves of different models for ICU admission outcome. In both (a) and (c) the black dotted lines indicate the baseline, representing predictions of a random classifier. (b) and (d) depict the calibration curves of the models for 30-day mortality and ICU admission outcomes, respectively. For the calibration curves the black dotted line represents the model ideal calibration.
