Abstract
Based on studies and experiments on the loss term of SVMs, we argue that 1-norm measurement is better than 2-norm measurement for outlier resistance. Thus, we modify the previous 2-norm soft margin smooth support vector machine (SSVM 2 ) to propose a new 1-norm soft margin smooth support vector machine (SSVM 1 ). Both SSVMs can be solved in primal form without a sophisticated optimization solver. We also propose a heuristic method for outlier filtering which costs little in training process and improves the ability of outlier resistance a lot. The experimental results show that SSVM 1 with outlier filtering heuristic performs well not only on the clean, but also the polluted synthetic and benchmark UCI datasets. © 2010 Springer-Verlag Berlin Heidelberg.