Skip to main content
. 2026 Jun 11;13:1830474. doi: 10.3389/fmed.2026.1830474

Figure 2.

Two side-by-side ROC curve charts compare five machine learning models’ performance in development (panel A) and external (panel B) validation settings, showing sensitivity versus 1-specificity. Area under curve values in the legends indicate that XGBoost performs best in development (AUC 0.964) and Logistic performs best in external validation (AUC 0.825), with all models performing better on development than external data.

Receiver operating characteristic (ROC) curves for model discrimination in the development and external validation cohorts. (A) ROC curves in the development cohort comparing logistic regression (AUC = 0.895), XGBoost (AUC = 0.964), decision tree (AUC = 0.915), support vector machine (AUC = 0.892), and naïve Bayes (AUC = 0.899). (B) ROC curves in the external cohort comparing logistic regression (AUC = 0.825), XGBoost (AUC = 0.808), decision tree (AUC = 0.758), support vector machine (AUC = 0.816), and naïve Bayes (AUC = 0.821). The gray diagonal line indicates no-discrimination performance (AUC = 0.5).