Skip to main content
. 2026 Aug 12;17:1907116. doi: 10.3389/fimmu.2026.1907116

Figure 4.

Side-by-side data visualization with two panels compares classification algorithms. Panel A; dot plot displays external-validation AUC with 95 percent confidence intervals for ten algorithms, ranking logistic regression, elastic-net, and LASSO logistic highest at 0.830. Panel B; horizontal bar chart shows overfitting gap, with logistic regression, elastic-net, and LASSO logistic lowest at 0.04, while LightGBM and XGBoost have the highest gap at 0.20.

Comparison of ten machine-learning algorithms: (A) external-validation AUC with 95% bootstrap confidence intervals; (B) training-to-external AUC gap, with algorithms ranked by external-validation AUC. Regularized logistic regression, LASSO logistic regression, and elastic-net achieved the joint-highest external-validation AUC, whereas tree-based ensembles showed the largest overfitting gaps.