An Approach for Feature Extraction and Selection To Detect Unsolicited Bulk Email " SPAM"

Abstract

The growing problem of unsolicited bulk email known as spam has generated an increasing need for reliable anti-spam filters. Filters of this type have so far been based mostly on manually constructed keyword patterns. Recently a Naïve Bayesian classifier has been trained to detect spam messages automatically. To improve the performance of the automated anti-spam filters this research introduces: 1- a new feature selection method, the Multi-Phase Feature Selection Method. 2- a new alternative feature weighting function 3- a simple classification algorithm, the Mean of the Feature Weighting Classification Algorithm. The introduced approaches are analyzed theoretically. Experiments were conducted using 1150 email messages to compare the new methods to previous published methods of Sahami et al. and I. Androutsopoulos et al. and the results were overall comparable.