Shown below are average cross-validation balanced error rates of UCI datasets (with 10 90:10 train validation splits). These are standardized datasets where each column in the training data is set to Euclidean length 1. The test data is normalized by the length on the training data points.

Dataset	Nearest means	Naive bayes	Least squares (on std data, eta=.001, term=.001)	Hinge loss (on std data, eta=.001, term=.001)
Breast cancer	9.3	7.9	5.5	5.0
Climate	16	17.5	9.1	9.1
Qsar	24.2	17.2	17.0	??
Ionosphere	24.7	53.6	15.8	16.9
Micromass	11.7	22.7	??	??
Hill valley (standardized)	46.8	48.2	34.4	50.2