Abstract:
The massive increase of spam is posing a very serious threat to email which has become an important means for communication. Not only it annoys users, but it also consumes much of the bandwidth of the Internet. Current spam filters are based on the contents of the email one way or the other. In this thesis we present a social network-based spam detection method in which the core idea is using social network measurements as feature to be used by classifier. Two separate classification models have been designed and tested. The first is k-Nearest-Neighbor Classifiers (KNN) classifier and the second is Locally weighted learning (LWL). The experimental results have shown a great favour of using KNN model for spam detection. However, it classifies many legitimate as spam which may annoy the email user. Hence we recommend this model to be applied where the acceptance of a spam message is more danger than legitimate messages rejection. While the classification result of LWL is better than KNN result. It is clear that KNN has advantage of detecting all spammer.
