Investigating the use of Bi-words for learning SPAM

Abstract

The target behind the artifact was to develop an anti-spam mail-client that would be able to classify e-mails according to whether they are spam or not. This was achieved by implementing two anti-spam algorithms that separately classified the e-mails in the best way possible, in order to avoid false-positive and false-negative e-mails. The results from the two anti-spam algorithms were quite satisfactory, especially that achieved by the first algorithm, which based it's classification on the occurrences of word-pairs in each e-mail. This algorithm achieved a maximum rate of correct ham classification of 92.6% and a maximum rate of correct spam classification of 98.4%. The second algorithm, which used only the bi-words which had low entropy, performed less well than algorithm one, especially when the system was not trained enough. The algorithms were left for the user's choice, from the application's main-window, that has the basic functionalities of popular mail-clients. The final windows-application, thoroughly showed the aims of the artifact.