Table 6.

Testing data results.

Model	Sensitivity	Specificity	Accuracy	Balanced accuracy	AUROCa
Ensemble-based DNNb with top 20 features	0.8324	0.8678	0.8400	0.8501	0.9216
Ensemble-based DNN with all 47 features	0.8160	0.8792	0.8307	0.8476	0.9211
DNN single model with top 20 features	0.8313	0.8662	0.8394	0.8488	0.9198
DNN single model with all 47 features	0.8345	0.8559	0.8395	0.8452	0.9184
SVMc	0.8359	0.8564	0.8407	0.8461	0.9204
LRd	0.8398	0.8482	0.8418	0.8440	0.9208
RFe	0.8463	0.8401	0.8448	0.8432	0.9187
LGBMf	0.8359	0.8531	0.8399	0.8445	0.9178
GBMg	0.8329	0.8548	0.8380	0.8438	0.9073
AdaBoosth	0.8393	0.8417	0.8399	0.8405	0.9169
XGBoosti	0.8373	0.8482	0.8399	0.8428	0.9172
Provisional diagnosis (clinicians)	1.000	0.000	0.7673	0.5000	N/Aj

^aAUROC: area under receiver operating characteristic.

^bDNN: deep neural network.

^cSVM: support vector machine.

^dLR: logistic regression.

^eRF: random forest.

^fLGBM: light gradient boosting machine.

^gGBM: gradient boosting machine.

^hAdaBoost: adaptive boosting.

ⁱXGBoost: extreme gradient boosting.

^jN/A: not applicable.