Spam filtering using Semantic and Rule Based model via supervised learning

Abstract

Exchanging information over internet becomes a mandatory requirement for everyone. Electronic mail is an important medium through which all types of information can be shared.  Despite the important role played by emails several challenges are also involved.  Email carries not only legitimate messages but several times unwanted messages also.  Spam email detection is challenging task since spammers use advanced technique through subject and email content. Emails with most fascinated words can be easily identified and classified as spam.  But the professional way in which spammers manage to send messages makes it difficult for researchers to classify. Always, Spammers concentrate on either subject or content or both to persuade users to open the email. In this paper, Rule Based Subject Analysis (RBSA) and Semantic Based Feature Selection (SBFS) techniques are integrated with the Machine Learning algorithms. Various rules are outlined to check the subject field of the emails.  Semantic based Feature selection technique is applied on the content of the email to reduce the features. RBSA and SBFS are integrated with four classifiers namely Support Vector Machine, Multinomial Naive Bayes, Gaussian Naive Bayes and Bernoulli Naive Bayes. The efficiency of the proposed techniques is tested on Enron dataset and it is observed that our proposed techniques with Support Vector Machine achieve lowest False Positive rate of 0.03.