Anti-spam Filters Based on Support Vector Machines

Recently, spam has become an increasingly important problem. In this paper, a support vector machine (SVM) is used as the spam filter. Then a study is made of the effect of classification error rate when different subsets of corpora are used, and of the filter accuracy when SVM's with linear, polynomial, or RBF kernels is used. Also an investigation is made of the effect of the size of attribute sets. Based on the experimental results and analysis, it is concluded that SVM will be a very good alternative for building anti-spam classifiers, with consideration of a good combination of accuracy, consistency, and speed.