Personalized spam filtering using incremental training of support vector machine

Continuously increasing ratio of spam mails has raised a serious issue regarding the Content of an e-mail and the user Consent for accepting an e-mail. The content of spam mails changes over a time period as, spammers apply different techniques to elude filters. The consent relates individual user's preferences for discriminating mails as spam or legitimate. We present personalized spam filter using incremental training of support vector machines (SVM). The filter is built with two different approaches. In the first approach, the filter is incrementally trained with support vectors and a set of incoming mails, keeping the same set of features. While in the second approach, the feature set is heuristically updated before applying incremental training to SVM. In depth comparison of results with conventional SVM batch training shows that the incremental approach achieves higher classification accuracy and lower false positive rate.