AUROC and AUPR curves.

The PR and ROC curves represent the performance of a binary classification model on bootstrapped test datasets with class imbalance (median curves with their AUCs are shown). a1 & a2 AUPR and AUROC curves of the gene expression MiRF model for OS ≥ 2 years and KPS ≥ 80 models, respectively. b1 & b2 AUPR and AUROC curves of the proteomics MiRF model for OS ≥ 2 years and KPS ≥ 80 models, respectively. c1 & c2 AUPR and AUROC curves of the integrated omics MiRF model for OS ≥ 2 years and KPS ≥ 80 models, respectively. An AUC value closer to 1 indicates better performance. In AUPR curves the color scale on the right side of the plot represents the value of the threshold. This threshold is used to calculate the precision and recall values for each point on the curve. Each shade represents a different threshold value. The fact that we have small data points in the test dataset is reflected in the PR curves which display only two colors (thresholds). This means precision and recall values at different classification thresholds may not be well represented. However, we coped with this problem by bootstrapping and taking the median of model performance, and through using a second parameter AUROC. This provides a comprehensive understanding of the model’s statistical performance on binary classification data with class imbalance.
(ZIP)