Table 2.

Performance of four different classifiers with three different feature selection methods in the modeling set

Classifier	Feature selection	AUC	Accuracy	Specificity	Sensitivity
LDA	WRST	0.67 ±0.03	0.70 ±0.02	0.81 ±0.11	0.53 ±0.07
	MRMR	0.64 ±0.03	0.67 ± 0.02	0.76 ±0.11	0.55 ± 0.07
	RF	0.63 ± 0.03	0.66 ±0.02	0.80±0.15	0.45 ± 0.09
QDA	WRST	0.54 ±0.04	0.63 ± 0.02	0.88 ±0.14	0.27 ± 0.08
	MRMR	0.56 ±0.03	0.64 ±0.02	0.86 ±0.11	0.31 ±0.07
	RF	0.54 ±0.03	0.62 ±0.02	0.89 ±0.13	0.24 ± 0.08
RF	WRST	0.63 ± 0.03	0.61 ±0.03	0.73 ±0.04	0.45 ± 0.04
	MRMR	0.66 ±0.04	0.63 ± 0.03	0.77 ± 0.06	0.44 ±0.04
	RF	0.63 ± 0.03	0.62 ±0.03	0.72 ±0.05	0.46 ± 0.04
SVM	WRST	0.67 ±0.02	0.64 ±0.02	0.85 ±0.06	0.33 ±0.03
	MRMR	0.66 ±0.04	0.63 ± 0.03	0.86 ±0.07	0.28 ±0.04
	RF	0.67 ±0.03	0.61 ±0.02	0.83 ±0.05	0.30 ±0.04

AUC area under the receptor operating curve, LDA/QDA linear/quadratic discriminant analysis, SVM support vector machine, RF Random Forest classifier, MRMR minimum redundancy, maximum relevance feature selection method, WRST Wilcoxon’s rank-sum test, RF Random Forest feature selection method

The best performance in each metric/column is shown in bold