Content‐Based Spam Email Classification using Machine‐Learning Algorithms
Eric P. Jiang · 2010
This chapter considers five supervised machine-learning algorithms for an evaluation study of spam filtering application. The algorithms includes: naive Bayes classifier (NB), support vector machines (SVMs), logitBoost algorithm (LB), augmented latent semantic indexing space model (LSI) and radial basis function (RBF) networks. Two benchmark email testing corpora for experiments that were constructed from two different languages and have reverse ratios of the number of spam emails to the number of legitimate emails in the training data were selected. It discusses several data preprocessing procedures, including feature selection and message representation. Spam filtering is a cost-sensitive classification task and a related discussion of effectiveness measures is included. The chapter compares the algorithms, using two popular email testing corpora. The experimental results and analysis are reported, and an empirical comparison of the characteristics of the five classifiers is presented. Controlled Vocabulary Terms latent class model; Multivariate statistics