AN IMPROVED OF SPAM E-MAIL CLASSIFICATION MECHANISM USING K-MEANS CLUSTERING

Abstract

Spam e-mails are considered as a serious violation of privacy. In addition, it has become costly and unwanted communication. Although, Support Vector Machine (SVM) has been widely used in e-mail spam detection, yet the problem of dealing with huge dat a is time and memory consuming and low accuracy. This study speeds up the computational time of SVM classifiers by reducing the number of support vectors. This is done by the K-means SVM (KSVM) algorithm proposed in this work. Furthermore, this paper proposes a mechanism for e-mail spam detection based on hybrid of SVM and K-means clustering and requires one more input parameter to be determined: the number of clusters. The experiment of the proposed mechanism was carried out using spambase standard dataset to evaluate the feasibility of the proposed method. The result of this hybrid method l ed to improved SVM classifier by reducing support vectors, increasing the accuracy and decreasing the time of e-mail spam detection. Experimental result s on spambase datasets showed that the improved SVM (KSVM) significantly outperforms SVM and many other recent spam detection methods in terms of cla ssification accuracy (effectiveness) and time consu ming (efficiency).