Table 1.

Comparison of test errors for support vector machine with linear kernel (SVM l), with polynomial kernel (SVM p), with sigmoid kernel (SVM s), with radial basis kernel (SVM r), RandomForest (RF), RandomForest with sparse variables removed (RFrm) and Supervised NMF

Dataset	SVM l	SVM p	SVM s	SVM r	RF	RFrm	Supervised NMF
Gut	0	0.2335	0	0.0661	0	0	0
Tongue	0.0202	0.2694	0.0202	0.0484	0.0081	0.0242	0.0040
Left Palm	0.1245	0.2691	0.1446	0.2691	0.1285	0.0643	0.0924
Right Palm	0.3455	0.2724	0.3455	0.1667	0.0732	0.0488	0.1789
Mammal	0.0714	0.1428	0.0714	0.1071	0.1429	0.1071	0
	[0.0461]	[0.0505]	[0.0461]	[0.0505]	[0.0505]	[0.0505]	[0]
Qin	0.3178	0.3359	0.2592	0.2853	0.2299	0.2299	0.2333
	[0.0567]	[0.0530]	[0.0516]	[0.0494]	[0.0573]	[0.0467]	[0.0515]

The first four rows are the prediction errors on the test data. The last two datasets are cross-validated errors with standard errors given in brackets on the line below. Best prediction for each dataset is presented in italics