Dynamic Concept Drift Detection for Spam Email Filtering

Abstract — Nowadays most of Internet users surfer from spam emails. Filtering technique is one of the effective methods which help us to get rid of the spam emails. One of the problems of filtering is that it cannot detect spam emails accurately when the concepts change or drift happens as time goes by. Therefore, it is required to handle concept drift accurately and quickly. This paper proposes a new algorithm for concept drift detection with three different levels; control, warning, and alarm level. The results show that the proposed algorithm can detect concept drift more accurately compared with the previously proposed ones. In addition, it can detect sudden concept changes more accurately. Index Terms — concept drift, content based filtering, machine learning, spam email, spam email detection I