Evolutionary Symbiotic Feature Selection for Email Spam Detection

Abstract

This work presents a symbiotic filtering approach enabling the exchange of relevant word features among different users in order to improve local anti-spam filters. The local spam filtering is based on a Content-Based Filtering strategy, where word frequencies are fed into a Naive Bayes learner. Several Evolutionary Algorithms are explored for feature selection, including the proposed symbiotic exchange of the most relevant features among different users. The experiments were conducted using a novel corpus based on the well known Enron datasets mixed with recent spam. The obtained results show that the symbiotic approach is competitive.