Phishing Emails Detection Using CS-SVM

Abstract

Phishing attacks are common online, which have resulted in financial losses through using either malware or social engineering. Thus, phishing email detection with high accuracy has been an issue of great interest. Machine learning-based detection methods, particularly Support Vector Machine (SVM), have been proved to be effective. However, the parameters of kernel method, whose default is that class numbers reciprocals in general, affect the classification accuracy of SVM. In order to improve the classification accuracy, this paper proposes a model, called Cuckoo Search SVM(CS-SVM). The CS-SVM extracts 23 features, which are used to construct the hybrid classifier. In the hybrid classifier, Cuckoo Search (CS) is integrated with SVM to optimize parameter selection of Radial Basis Function(RBF). Experiments are performed on a dataset consisting of 1,384 phishing emails and 20,071 non-phishing emails. Experimental results show that the proposed method has higher phishing email detection accuracy than SVM classifier with default parameter value. The CS-SVM classifier can obtain the highest accuracy of 99.52 percent.