Filtering SPAM Using Several Stages Neural Networks

Many unsolicited commercial emails (SPAM) are delivered to email users' mailboxes, which may cause to lose important emails and wasting user's time on deleting SPAM. This paper applied Several Stage Neural Networks method to filter SPAM. This method has shown a good performance comparable to other used methods, using less computational resources. Moreover its performance on filtering SPAM was better than its performance in scenes classification, where originally it was applied. The tests were done with two types of neural networks: Feed-forward and Self-organizing Global Ranking Map, which was trained in two ways: expanding input vector and training on each class separately. The results showed that feed-forward is suitable for classifying emails having single or few subjects or realms, and self-organizing global ranking map is suitable for classifying emails with many realms.