Table 4.

Comparison and prediction performances of different models for LNM.

Models	AUC (95% CI)	Accuracy	Precision	F1-score	Recall-rate	Specificity
Training set
XGBoost	0.791 (0.776–0.806)	0.739	0.690	0.663	0.638	0.894
SVM	0.733 (0.707–0.750)	0.732	0.696	0.640	0.593	0.944
KNN	0.719 (0.702–0.737)	0.738	0.727	0.653	0.592	0.961
LR	0.747 (0.733–0.763)	0.728	0.671	0.646	0.623	0.888
RF	0.772 (0.757–0.787)	0.734	0.713	0.645	0.589	0.956
LightGBM	0.772 (0.757–0.787)	0.736	0.691	0.653	0.619	0.916
Test set
XGBoost	0.829 (0.818–0.843)	0.770	0.738	0.706	0.677	0.912
SVM	0.791 (0.778–0.805)	0.755	0.755	0.681	0.620	0.960
KNN	0.634 (0.618–0.653)	0.715	0.666	0.607	0.558	0.955
LR	0.795 (0.780–0.808)	0.748	0.706	0.675	0.647	0.905
RF	0.821 (0.808–0.833)	0.740	0.745	0.659	0.591	0.969
LightGBM	0.826 (0.813–0.839)	0.759	0.729	0.687	0.650	0.925

Abbreviations: XGBoost, extreme gradient boosting; SVM, support vector machine; KNN, k-nearest neighbor; LR, logistic regression; RF, random forest; LightGBM, light gradient boosting machine; LNM, lymph node metastasis.