A generalized additive neural network application in information security

Abstract

Traditionally spam has been considered as an inconvenience requiring workers to sift through and delete large numbers of e-mail messages per day. However, new developments and the Internet have dramatically transformed the world and over the last number of years a situation has been reached where inboxes have been flooded with unsolicited messages. This has caused spam to evolve into a serious security risk with prominent threats such as spreading of viruses, server problems, productivity threats, hacking and phishing etc. To combat these and other related threats, efficient security controls such as spam filters, should be implemented. In this paper the use of a Generalized Additive Neural Network (GANN), as a spam filter, is investigated. A GANN is a novel neural network implementation of a Generalized Additive Model and offers a number of advantages compared to neural networks in general. The performance of the GANN is assessed on three publicly available spam corpora and results, based on a specific classification performance measure, are presented. The results showed that the GANN classifier produces very accurate results and may outperform other techniques in the literature by a large margin.