Table 6.

The four best predictive models, all chosen by NCA feature selection. Accuracy, AUC, sensitivity, specificity, precision, and F-score are shown separately for each class. The best classifiers were the cosine KNN, fine KNN, subspace KNN, and random forest. The best performances for each class are shown in bold

	Classifiers	Classes	Accuracy (mean ± SD)	AUC (mean ± SD)	Sensitivity (mean ± SD)	Specificity (mean ± SD)	Precision (mean ± SD)	F-score (mean ± SD)
NCA	Cosine KNN	Class 1	0.74 ± 0.033	0.83 ± 0.083	0.76 ± 0.047	0.72 ± 0.041	0.71 ± 0.035	0.73 ± 0.036
		Class 2	0.87 ± 0.023	0.92 ± 0.092	0.70 ± 0.067	0.92 ± 0.018	0.72 ± 0.052	0.71 ± 0.053
		Class 3	0.84 ± 0.024	0.85 ± 0.085	0.69 ± 0.045	0.91 ± 0.028	0.78 ± 0.055	0.73 ± 0.039
	Fine KNN	Class 1	0.79 ± 0.029	0.81 ± 0.080	0.82 ± 0.045	0.76 ± 0.039	0.75 ± 0.032	0.78 ± 0.031
		Class 2	0.86 ± 0.022	0.84 ± 0.084	0.77 ± 0.051	0.89 ± 0.027	0.68 ± 0.052	0.72 ± 0.038
		Class 3	0.87 ± 0.021	0.83 ± 0.083	0.67 ± 0.052	0.97 ± 0.020	0.90 ± 0.054	0.77 ± 0.044
	Subspace KNN	Class 1	0.72 ± 0.034	0.82 ± 0.081	0.79 ± 0.043	0.66 ± 0.053	0.67 ± 0.036	0.72 ± 0.032
		Class 2	0.84 ± 0.022	0.88 ± 0.088	0.70 ± 0.064	0.88 ± 0.020	0.63 ± 0.046	0.66 ± 0.047
		Class 3	0.85 ± 0.026	0.83 ± 0.083	0.57 ± 0.073	0.97 ± 0.025	0.91 ± 0.078	0.69 ± 0.060
	Random forest	Class 1	0.76 ± 0.037	0.89 ± 0.089	0.81 ± 0.051	0.71 ± 0.059	0.71 ± 0.043	0.75 ± 0.035
		Class 2	0.82 ± 0.037	0.90 ± 0.090	0.51 ± 0.012	0.91 ± 0.032	0.63 ± 0.010	0.56 ± 0.010
		Class 3	0.78 ± 0.039	0.84 ± 0.084	0.60 ± 0.080	0.86 ± 0.046	0.66 ± 0.084	0.63 ± 0.066