On Improving the Performance of Spam Filters Using Heuristic Feature Selection Techniques

Electronic mail (E-mail) has become extremely important in our daily life because of its high speed and low cost. Unfortunately, every day, E-mail users receive increasing number of unwanted spam E-mails from different sources. Spam E-mail has become an annoying, costly and time-consuming problem for many people. In fact, spam E-mail has become one of the most common ISP customer service complaints and one of the main reasons behind most subscriber churn. One popular means for solving the spam problem is to deploy an email filter to classify the spam and legitimate E-mails. However, the accuracy of most of the current solutions still needs further improvement. In this paper, we present two heuristic feature selection algorithms that can be used to improve the accuracy of spam email filters. In particular, we experiment the application of both the artificial immune systems (AIS) and tabu search (TS) as classifier dependent feature selection techniques for email filter. We also compare the performance of our proposed solution with the classical singular value decomposition (SVD) based system. Using a K-nearest neighbor (KNN) classifier, the accuracy of the AIS and the tabu search based systems is 90.9% and 94.5% respectively as compared to 90% for the SVD based system