Table 3.

RNA-Seq-based BRCA subtypes classification using 5-fold cross-validation with 100 repeats. The first column denotes the five kinds of subtypes, and we built a binary classifier for each subtype by splitting the data into control and experiment groups. The sample size of two groups was imbalanced, so the “SMOTE” sampling method in the second column was utilized to lessen the interference of imbalanced data. The “LumA” subtype was an exception because it had sufficient samples. The third column denotes the five kinds of metrics used in this experiment, and the remaining columns are the three kinds of machine learning approaches adopted in this research, where the “svmRadial” represents the svm with radial basis kernel.

Subtypes	Sampling	Metrics	nb	rf	svmRadial
Basal-like	SMOTE	Sensitivity	0.9737	0.9605	0.9737
		Specificity	0.9580	0.9916	0.9720
		Accuracy	0.9607	0.9861	0.9723
		F1	0.8970	0.9605	0.9250
		AUC	0.9847	0.9976	0.9968
Her2	SMOTE	Sensitivity	0.9063	0.7813	0.8750
		Specificity	0.8853	0.9601	0.9526
		Accuracy	0.8868	0.9469	0.9469
		F1	0.5421	0.6849	0.7089
		AUC	0.9562	0.9797	0.9798
LumA	None	Sensitivity	0.9067	0.8667	0.9067
		Specificity	0.8173	0.8846	0.8462
		Accuracy	0.8637	0.8753	0.8776
		F1	0.8737	0.8784	0.8850
		AUC	0.9134	0.9952	0.9481
LumB	SMOTE	Sensitivity	0.8415	0.8171	0.5488
		Specificity	0.8376	0.9288	0.9544
		Accuracy	0.8383	0.9076	0.8776
		F1	0.6635	0.7701	0.6294
		AUC	0.9075	0.9494	0.9043
Normal-like	SMOTE	Sensitivity	0.8125	0.7500	0.5000
		Specificity	0.8517	0.9498	0.9833
		Accuracy	0.8502	0.9424	0.9654
		F1	0.9163	0.9695	0.9821
		AUC	0.9125	0.9600	0.9640