A Collaborative Abstraction Based Email Spam Filtering with Fingerprints

Spam detection in emails tends to be an endless research interest among many researchers and academicians. Even though email communication has become a major role in day to day activities, the increasing volumes of threats towards spam emails has paved the way for numerous email spam detection techniques. Many spam filtering methods including data mining and machine learning techniques are adopted by researchers; yet a complete accurate filtering model is an expected solution to cope up with the intentional spam attacks. This paper proposes one such model that uses a hybrid approach towards efficient spam detection. A collaborative spam filtering framework using abstraction of the entire email layout and the fingerprints of the layout is proposed to match and catch the sprouting nature of spam. Collaborative framework uses recommendations from other users to create spam database. Any incoming mail is checked against the spam database for spam or ham classification using near duplicate similarity matching scheme. To reduce false positive and false negative ratio in spam classification, we calculate cumulative weights from both email layouts and fingerprints. Fingerprint signatures of newly classified spam are progressively updated to the spam database for up-to-date spam detection. The system is evaluated with Spam Assassin dataset and the results are proven for a comparatively better performance.

Automated summary

This paper proposes a hybrid approach for collaborative spam detection, which abstracts the entire email layout and extracts layout fingerprints to effectively match and catch the sprouting nature of spam. The system creates a spam database using recommendations from other users, calculates cumulative weights to reduce false positive and false negative ratio, and progressively updates the fingerprints of newly classified spam for up-to-date spam detection. The system is evaluated with the Spam Assassin dataset and shows comparatively better performance.