Construction of Xinjiang metabolic syndrome risk prediction model based on interpretable models

BMC Public Health

Table 2 Classification performance comparison between the DT and LR models

Model	Datasets	Accuracy	Sensitivity	Specificity	AUROC (95% Cl)
LR
Model 1	Original imbalanced	0.901	0.418	0.971	0.694(0.684 ~ 0.704)
Model 2	Random oversampling	0.839	0.854	0.837	0.846^*(0.837 ~ 0.854)
Model 3	Random undersampling	0.843	0.851	0.842	0.846^*(0.838 ~ 0.854)
Model 4	Hybrid sampling	0.839	0.856	0.837	0.847^*(0.838 ~ 0.855)
Model 5	SMOTE	0.838	0.855	0.836	0.846^*(0.837 ~ 0.854)
DT
Model 6	Original imbalance	0.915	0.588	0.962	0.775(0.766 ~ 0.785)
Model 7	Random oversampling	0.874	0.942	0.864	0.903^#(0.896 ~ 0.910)
Model 8	Random undersampling	0.879	0.959	0.868	0.913^#(0.907 ~ 0.919)
Model 9	Hybrid sampling	0.873	0.926	0.866	0.896^#(0.889 ~ 0.902)
Model 10	SMOTE	0.851	0.920	0.841	0.880^#(0.873 ~ 0.888)

* P <0.05 compared with the AUROC value of Model 1. # P <0.05 compared with the AUROC value of Model 6.

ISSN: 1471-2458