New hybrid features for phish website prediction

Abstract

Phishing is a serious threat to the web economy and the Internet communication, because phishers put both users and organizations at risk of identity theft and financial losses. Phishers continually exploit new sophisticated features to impersonate legitimate web pages, modify their components and host their phishes. Furthermore, the prediction susceptibilities of features that were previously investigated become a key challenge for discriminating the evolving phishes. Accordingly, this paper investigated the prediction susceptibility of 58 hybrid features. It was observed that the investigated features were highly exploited in the content and hosted the URLs of phish webpages. The prediction susceptibility of the proposed features was experimentally examined in the suspected webpages using the SVM machine learning classification technique. The results revealed that the introduced features could be considered as potentially predictive ones and they could be utilized in the upcoming research to improve phishing detection approaches.