Email classifier: An ensemble using probability and rules

On a low cost and effective communication via internet using emails, spam mails have come up as a dark spot. Many researchers have proposed a solution to this spam problem by using one classifier or combining more than one classifier. The latter has proved to be much more efficient than the individual classifiers. In this paper, we propose an elementary classifier combination, diversified both by features set and different classifiers. The proposed ensemble combines multiple classifiers in four levels such that a test set given to each classifier depends on the previous level´s classifier results. Also, the proposed scheme uses meta-learning technique. The ultimate decision is made using the classifiers prediction, their probability of prediction and some combining rules to classify legitimate and spam mails more precisely. We evaluate the performance of our scheme in terms of accuracy, precision, recall, F1 score and ROC curve. All of these performance measure shows that our scheme is more accurate than individual classifiers.