Behavior-based spam detection using a hybrid method of rule-based techniques and neural networks

Earlier methods on spam filtering usually compare the contents of emails against specific keywords, which are not robust as the spammers frequently change the terms used in emails. This paper presents a hybrid method of rule-based processing and back-propagation neural networks for spam filtering. Instead of using keywords, this study utilize the spamming behaviors as features for describing emails. A rule-based process is first employed to identify and digitize the spamming behaviors observed from the headers and syslogs of emails. An enhanced BPNN with a weighted learning strategy is designed as the classification mechanism. Since spamming behaviors are infrequently changed, compared with that of keywords used in spams, the proposed method is more robust with respect to the change of time. The experimental results show that the proposed method is useful in identifying spam emails.