Table 4.

Summary of the predictive performance of each prediction model.

Model	AUROC	1-Accuracy	Sensitivity	Specificity	MCC
Variable selection by logistic regression using backward elimination *
Logistic Regression (Final model)	0.784	0.235	0.799	0.531	0.258
XGBoost	0.785	0.237	0.769	0.581	0.148
Random Forest	0.780	0.245	0.759	0.400	0.040
LightGBM	0.774	0.236	0.782	0.546	0.204
Variable Selection by random forest using permutation importance **
Logistic Regression	0.783	0.232	0.798	0.547	0.260
XGBoost	0.782	0.237	0.770	0.583	0.158
Random Forest	0.779	0.245	0.772	0.480	0.139
LightGBM	0.780	0.239	0.776	0.532	0.174

Analyses were performed with test dataset (n = 1259). AUROC: area under the receiver operating characteristic; 1-Accuracy: misclassification rate; MCC: Matthews correlations coefficient. * Five variables were included: age, household income, daily brushing frequency, age of the mother at giving birth, and the mother’s DMFT quartile. ** Four variables were included: age, household income, daily brushing frequency, and the mother’s DMFT quartile.