home - SVM classification

Classification results

We sample 1000 examples per model for each real data set, train a pairwise multi-class SVM on 4/5 of the sampled data and test on a 1/5 hold-out set. We determine a prediction by counting votes for the different classes.

	E. coli	C. elegans	S. cerevisiae
average training-loss	1.6%	0.5%	2.1%
average test-loss	1.6%	0.5%	1.8%
average number of support vectors	109	51	106
winning model	Kumar	Middendorf-Ziv	Sole
robustness	1.0	0.97	0.64

• Detailed SVM results: E. coli, C. elegans, S. cerevisiae