An Artificial Neural Nets for Spam e-mail Recognition

Abstract

The volume of unsolicited commercial e-mail messages transmitted by the Internet has reached epidemic proportions nowadays. As a computer viruses spam is changing all time. Spam gets not only new forms every day but new exploit techniques also. The recognition capability of the ANN is found to be good, but because of low but nonzero false spam recognition the ANN is not suitable for use alone as a spam rejection tool. In fact, false identification of legitimate email is worse than receiving spam message, so the filter that yields false positive is not suitable. From other hand there is no sense to increase the text parameters number, because the difference of classification precision considering false positive value between 41 and 57 input nets is minimal. The ANN, unlike the statistical classifier, is possible to train with additional input parameters related not only to text, but implementing pattern recognition also, especially when most today’s spam messages are not even included with unsolicited text which allows quickly recognize spam using ANN identifying keywords, but uses graphical media as attachment to normal text e-mail. Ill. 3, bibl. 9 (in English; summaries in English, Russian and Lithuanian).