Table 2.

Model evaluation.

Model	AUROC	Accuracy, %	PPV, %	NPV, %	BA, %	F1-score	MCC
Internal validation
LightGBM	0.807	76.9	60.1	81.6	67.9	0.533	0.387
CatBoost	0.795	76.2	60.1	79.7	64.8	0.478	0.344
GBDT	0.798	76.0	60.3	79.4	64.2	0.333	0.336
XGBoost	0.780	75.3	56.3	80.7	66.1	0.414	0.345
Random Forest	0.792	76.3	61.2	79.5	64.5	0.326	0.343
LR	0.758	75.1	60.6	77.1	60.0	0.258	0.275
Bagging	0.761	74.2	54.7	78.8	62.7	0.384	0.291
SVM	0.757	72.4	45.5	76.5	56.4	0.351	0.169
Naïve Bayes	0.640	72.7	51.1	75.3	56.3	0.274	0.182
MLP	0.651	69.8	43.3	76.5	58.0	0.456	0.184
KNN	0.608	71.6	42.7	73.5	52.4	0.211	0.089
AdaBoost	0.790	74.9	55.7	80.0	64.9	0.404	0.326
Decision Tree	0.603	69.0	43.8	78.9	61.6	0.536	0.230
External validation
LightGBM	0.816	82.8	59.0	85.0	60.9	0.362	0.311
CatBoost	0.791	82.6	62.4	83.7	57.5	0.271	0.262
GBDT	0.778	82.1	62.1	82.7	54.2	0.171	0.195
XGBoost	0.790	81.9	52.9	85.4	62.0	0.384	0.303
Random Forest	0.804	82.4	58.7	84.0	58.2	0.293	0.264
Logistic Regression	0.755	82.0	55.9	83.6	57.0	0.261	0.234
Bagging	0.760	81.7	52.1	84.4	59.3	0.325	0.261
SVM	0.707	81.5	46.9	83.6	55.6	0.227	0.185
Naïve Bayes	0.569	73.0	29.2	85.3	58.5	0.322	0.156
MLP	0.692	80.9	47.2	83.6	57.0	0.268	0.207
KNN	0.606	81.4	57.1	81.6	50.9	0.042	0.084
AdaBoost	0.793	82.2	54.4	85.0	61.0	0.363	0.294
Decision Tree	0.622	74.8	34.5	85.3	61.1	0.367	0.212

The bold values represent the best predict performance among these models. AUROC, the area under the receiver operating characteristic curve; PPV, positive prediction value; NPV, negative prediction value; BA, balanced accuracy; MCC, Matthews correlation coefficient.