SpamNet -Spam Detection Using PCA and Neural Networks

摘要

This paper describes SpamNet - a spam detection program, which uses a combination of heuristic rules and mail content analysis to detect and filter out even the most cleverly written spam mails from the user's mail box, using a feed-forward neural network. SpamNet is able to adapt itself to changing mail patterns of the user. We demonstrate the power of Principal Component Analysis to improve the performance and efficiency of the spam detection process, and compare it with directly using words as features for classification. Emphasis is laid on the effect of domain specific preprocessing on the error rates of the classifier.