Spam email classifcation with generalized additive neural networks using ensemble methods

Abstract

Solutions to minimize the amount of junk mail we receive are not yet optimal and thus research towards an improved spam email classifier is ongoing. Traditional content and listbased filters are less effective than machine learning approaches, since continuous maintenance and human input are necessary. In this paper an automated construction algorithm for a Generalized Additive Neural Network (GANN) is applied to the Enron and PU1 email corpora and compared to the popular Naive Bayesian filter. In an attempt to further enhance the GANN performance, both Bagging and Boosting ensemble methods were applied. It was found that both ensemble methods improved results making the GANN the better choice