A study of machine learning classifiers for spam detection

Abstract

In the present world, there is a need of emails communication but unsolicited emails hamper such communications. The present research emphasises to build a spam classification model with/without the use of ensemble of classifiers methods have been incorporated. Through this study, the aim is to distinguish between ham emails and spam emails by making an efficient and sensitive classification model that gives good accuracy with low false positive rate. Greedy Stepwise feature search method has been incorporated for searching informative feature of the Enron email dataset. The comparison has been done among different machine learning classifiers (such as Bayesian, Naïve Bayes, SVM (support vector machine), J48 (decision tree), Bayesian with Adaboost, Naïve Bayes with Adaboost). The concerned classifiers are tested and evaluated on metric (such as F-measure (accuracy), False Positive Rate, and training time). By analysing all these aspects in their entirety, it has been found that SVM is the best classifier to be used. It has the high accuracy and the low false positive rate. However, training time of SVM to build the model is high, but as the results on other parameters are positive, the time does not pose such an issue.