Learning to filter spam emails: An ensemble learning approach

Article type: Research Article

Authors: Saeedian, Mehrnoush Famil | Beigy, Hamid

Affiliations: Department of Computer Engineering, Sharif University of Technology, Tehran, Iran. E-mail: saeedian@ce.sharif.edu, beigy@ce.sharif.edu

Abstract: Most email users have experienced spam problems, which have been addressed as text classification problem. In this paper, we propose a novel spam detection method which uses an ensemble of classifiers based on subsampling and dynamic weighted voting techniques. Since there is diversity in genre of emails' contents, the proposed method finds different topics in emails by using a clustering algorithm. The proposed algorithm first extracts disjoint clusters of emails, and then a classifier is trained on each cluster, and finally decisions of classifiers are combined using dynamic weighted majority techniques. In order to classify a new input sample, first it is compared with all cluster centers and its similarity to each cluster is identified; then the classifiers in the vicinity of the input sample obtain greater weights for the final decision of the ensemble. Finally, the outputs of the classifiers are combined using weighted voting with weights calculated from the similarity of the input sample with cluster centers. The experimental results show that the proposed algorithm outperforms pure SVM and the related ensemble based classifiers.

Keywords: Spam detection, content based spam filtering, machine learning, ensemble learning

DOI: 10.3233/HIS-2011-0145

Journal: International Journal of Hybrid Intelligent Systems, vol. 9, no. 1, pp. 27-43, 2012