. 2015 Apr 2;4(2):162–187. doi: 10.3390/microarrays4020162

Table 5.

Performance of classifier panels calculated with class prediction analysis. Different normalization strategies were used and different subsets of samples were analyzed. The analysis of all samples together means that all four histological types of lung cancer were merged to one cancer class.

Analyzed sample set	Data normalization	Feature selection^a	Classifier size	Best predictor^b	CC^c	Sensitivity	Specificity
All cases vs. controls (run 1–4)	QN	50 GP	100	1-NN, SVM	83%	0.90	0.75
All cases vs. controls (run 1–6)	QN	100 RFE	100	3-NN	79%	0.76	0.83
All cases vs. controls (run 1–4)	ComBat	25 GP	50	3-NN	81%	0.85	0.78
All cases vs. controls (run 1–6)	ComBat	25 GP	50	SVM	85%	0.85	0.85
SCLC vs. control group (run 1 + 5)	QN	100 RFE	100	SVM	98%	1.00	0.96
	ComBat			SVM	94%	0.92	0.96
	DWD			SVM	92%	0.88	0.96
SqLC vs. control group (run 2 + 5)	QN	100 RFE	100	SVM	88%	0.80	0.96
	ComBat			CCP, NC, SVM	96%	0.96	0.96
	DWD			SVM	81%	0.76	0.87
LCLC vs. control group (run 3 + 6)	QN	100 RFE	100	CCP, SVM	85%	0.84	0.87
	ComBat			1-NN	92%	0.88	0.96
	DWD			SVM	79%	0.72	0.87
AdCa vs. control group (run 4 + 6)	QN	100 RFE	100	SVM	89%	0.92	0.87
	ComBat			CCP, DLDA, NC	83%	0.83	0.83
	DWD			DLDA	91%	0.92	0.91

^a greedy-pairs algorithm (GP), recursive feature elimination (RFE); ^b Nearest Neighbor classification (NN), support vector machine(SVM), Compound Covariate Predictor (CCP), Nearest Centroid (NC), Diagonal Linear Discriminant Analysis (DLDA); ^c correct classification rate.