SMS Spam filtering using N-gram method, information gain metric and an improved version of SVDD classifier

Abstract

Text messages (SMS or Short Message Service) are widely used form of mobile communication; their popularity is attributed to many factors including low cost sending, simple delivery mode and convenient usage. However, an unsolicited kind of SMS (Spam) has been appeared and caused major problems for users and mobile service providers. In this paper we propose a new SMS Spam filter able to distinguish between legitimate messages and Spam. The proposed filter is based on three components: N-grams method to extract features from short messages, information gain ratio to select the most relevant features, and an improved version of Support Vector Domain Description to detect SMS Spam. Experimental results on a large benchmark dataset of real-world benign and Spam SMS have shown the performance and effectiveness of the proposed filter.