Table 3. The performance of Support vector machine (SVM) and Random Forest (RF) based models developed using different sets of selected features on training and independent or external validation dataset.

Features	Dataset	Technique	Performance Measures
			Sensitivity	Specificity	Accuracy (%)	MCC	ROC
setA-1 (4 genes)	Training	SVM	71.65	70.3	71.12	0.41	0.76
	Validation		68.25	78.05	72.12	0.45	0.80
	Training	RF	70.87	65.45	68.74	0.36	0.69
	Validation		73.02	58.54	67.31	0.32	0.74
setB-1 (4 genes)	Training	SVM	71.26	70.3	70.88	0.41	0.74
	Validation		74.6	68.29	72.12	0.42	0.74
	Training	RF	80.31	49.7	68.26	0.32	0.65
	Validation		82.54	51.22	70.19	0.36	0.68
Combo-1 (8 genes)	Training	SVM	75.20	70.30	73.27	0.45	0.77
	Validation		77.78	68.29	74.04	0.46	0.80
	Training	RF	81.1	55.15	70.88	0.38	0.73
	Validation		82.54	51.22	70.19	0.36	0.74

These gene sets include setA-1 (4 overexpressed genes), setB-1 (4 under-expressed genes) and Combo-1 (combination of both gene sets i.e. setA-1 and setB-1).