Interplay between Probabilistic Classifiers and Boosting Algorithms for Detecting Complex Unsolicited Emails

This paper presents the performance comparison of probabilistic classifiers with/without the help of various boosting algorithms, in the Email Spam classification domain. Our focus is on complex Emails, where most of the existing classifiers fail to identify unsolicited Emails. In this paper we consider two probabilistic algorithms i.e. and Naive Bayes and three boosting algorithms i.e. Bagging, with Re-sampling and AdaBoost. Initially, the classifiers were tested on the Enron Dataset without Boosting and thereafter, with the help of Boosting algorithms. The Genetic Search Method was used for selecting the most informative 375 features out of 1359 features created at the outset. The results show that, in identifying complex Spam massages, Bayesian classifier performs better than Naive Bayes with or without boosting. Amongst boosting algorithms, with Resample has brought significant performance improvement to the Probabilistic classifiers.