Hybrid features-based prediction for novel phish websites

Abstract

Phishers frequently craft novel deceptions on their websites and circumvent existing anti-phishing techniques for insecure intrusions, users’ digital identity theft, and then illegal profits. This raises the needs to incorporate new features for detecting novel phish websites and optimizing the existing anti-phishing techniques. In this light, 58 new hybrid features were proposed in this paper and their prediction susceptibilities were evaluated by using feature co-occurrence criterion and a baseline machine learning algorithm. Empirical test and analysis showed the significant outcomes of the proposed features on detection performance. As a result, the most influential features are identified, and new insights are offered for further detection improvement.