Table 2.

Summary of AUC of the ROCs of the 3 Machine Learning Classifiers Under 3 Sampling Strategies

Method	Without Oversampling	Retinal Oversampling	Borderline-SMOTE Oversampling
	AUC	AUC	AUC
LDA	0.863 ± 0.015	0.958 ± 0.003	0.885 ± 0.011
SVM	0.859 ± 0.019	0.967 ± 0.008	0.975 ± 0.003
RF	0.880 ± 0.019	0.981 ± 0.006	0.986 ± 0.007

The area under the receiver operating characteristic curve (AUC of the ROC) is shown ± standard deviation for each of the 3 classifiers: linear discriminant analysis (LDA), support vector machine (SVM) and random forest (RF) without and with 2 oversampling methods, retinal oversampling and borderline-SMOTE oversampling. The corresponding curves are shown in Figure 4.