Phishing Classification Models an Empirical Study of Induction Factors for Effective Classification

摘要

Recently, researchers have devoted prominent machine learning-based anti-phishing models to survive a supreme cyber-security versus phishing evolution on the cyberspace. Yet, such models remain incompetent to detect new phish in a real-time application. In this concern, this paper advocates an empirical analysis with the recently published works via a chronological validation. Chronological validation achieved by testing the works on three benchmarking data sets to appraise the causality between their detection outcomes and their limitations. Throughout chronological validation, the tested works have fallen short at detecting new phish web pages with an accessible detection accuracy. High to moderate faults and misclassifications are resulted as implications for their limitations and fixed real-time settings. Accordingly, this paper infers that by elevating the tested models in terms of using new and hybrid features, robust subset of features, and actively learned classifiers; an adaptive anti-phishing model with adjustable settings will be resilient against the up-to-date and scalable web flows. With such inferences, this paper highlights what future trends to develop along with depicting a taxonomy of current status and open problems as a guide to the researchers for their future achievements.