Prediction of 3-year risk of diabetic kidney disease using machine learning based on electronic medical records

Table 2 Performance of the prediction models generated by the seven machine learning algorithms

Models	AUC	95% CI		SE (recall)	SP	AC	F1	PPV	NPV
Models	AUC	Lower bound	Upper bound	SE (recall)	SP	AC	F1	PPV	NPV
LightGBM	0.815	0.747	0.882	0.741	0.797	0.768	0.768	0.797	0.741
XGBoost	0.779	0.706	0.853	0.682	0.785	0.732	0.725	0.773	0.697
AdaBoost	0.805	0.738	0.872	0.659	0.772	0.713	0.704	0.757	0.678
Artificial Neural Network	0.800	0.730	0.869	0.659	0.911	0.768	0.747	0.862	0.680
Decision Tree	0.579	0.503	0.655	0.576	0.595	0.579	0.587	0.598	0.603
Support Vector Machine	0.791	0.720	0.862	0.612	0.886	0.744	0.712	0.852	0.680
Logistic Regression	0.798	0.728	0.868	0.718	0.759	0.738	0.739	0.763	0.714

SE: sensitivity; SP: specificity; AC: accuracy; PPV: positive predictive value; NPV: negative predictive value

ISSN: 1479-5876