Set Intersection Clustering for Spam Detection in Emails

Emails play a very important role in the business and social world. At the same time, there is a large amount of unsolicited emails, or spam, which contain malicious content such as malware and phishing. The increase in the volume of messages and the greater complexity in spam techniques make it necessary to adopt efficient methods to identify it. In this regard, for classifying emails as spam, this work proposes a model entitled Intersect Bayes. The proposed model combines Bayes' Theorem with the Set Intersection Clustering Problem (SICP) to explore the dependencies between words in a labeled emails database. To this end, we introduce a new mathematical formulation for SICP and a polynomial heuristic to obtain approximate solutions. The results demonstrated that the proposed model presents better classification metrics compared to classical models in the literature based on Bayes' Theorem, with a gain of up to 6% in accuracy.