A novel approach toward spam detection based on iterative patterns

Abstract

Spamming is becoming a major threat that negatively impacts the usability of e-mail. Although lots of techniques have been proposed for detecting and blocking spam messages, Spammers still spread spam e-mails for different purposes such as advertising, phishing, adult and other purposes and there is not any complete solution for this problem. In this work we present a novel solution toward spam filtering by using a new set of features for classification models. These features are the sequential unique and closed patterns which are extracted from the content of messages. After applying a term selection method, we show that these features have good performance in classifying spam messages from legitimate messages. The achieved results on 6 different datasets show the effectiveness of our proposed method compared to close similar methods. We outperform the accuracy near +2% compared to related state of arts. In addition our method is resilient against injecting irrelevant and bothersome words.