Table 4. Performance metrics of classifiers on the lung cancer test set (subtype and stage classification).

	N# of Genes	Error (%)	GBS (0)	BCM (1)	AUPR (1)
A. On classifiers
Ben-Hamo's study	23	49.3	NA	0.48	0.46
Hierarchical-TGDR	66	53.3	0.3736	0.4401	0.4709
Pairwise Coupling	158	54	0.3794	0.4371	0.4010
Multi-TGDR local	83	54	0.3579	0.4210	0.4681
Multi-TGDR global	60	54	0.3524	0.4164	0.4685
B. Radviz on 8-gene signature selected by subtype and stage classification
Radviz + multi-TGDR	8	54.7	0.3423	0.4137	0.4557
Radviz+ naïve Bayes	8	54.7	0.4104	0.4437	0.4494
Radviz+SVM	8	54	0.3654	0.4137	0.4562
C. Radviz on the most frequently selected features
Radviz + multi-TGDR	10	53.3	0.3215	0.4269	0.4710
Radviz+ naïve Bayes	10	55.3	0.4256	0.4503	0.4573
Radviz+ SVM	10	54.7	0.3516	0.3612	0.4815

NA: not available.