Authors: Hiba Zuhair; Ali Selamat
Addresses: Department of Systems Engineering, College of Information Engineering, Al-Nahrain University, Baghdad, Iraq ' Centre for Information and Communication Technologies, Software Engineering Department, Faculty of Computing, Universiti Teknologi Malaysia (UTM), Johor, Malaysia
Abstract: The never-ending threats of phishing to the cyberspace motivate researchers to develop more proficient phishing classification models for the survival of cyber-security with safe web services. However, these models remain variable in their reaction and incompetent in their performance against novel phishes at the real-time of application. This attributed to their partial or full deficiency of inductive factors including a rich set of decisive features, actively learned and adaptive machine learning based classification model. Upon this issue, our paper revisits the current machine learning-based phishing classification models and analyses their performance qualitatively and quantitatively across three benchmarking data sets. Empirical results and observations emphasised the causality between the models' limitations and their lack of inductive factors. Accordingly, future outlooks are recommended as a navigating taxonomy to serve the researchers at developing their upcoming achievements in both academia and industry.
Keywords: novel phish; phishing classification model; machine learning; feature-based classifier; FBC; ensemble feature-based classifier; EFBC; inductive factors; active learning; adaptive model.
International Journal of Digital Enterprise Technology, 2019 Vol.1 No.3, pp.219 - 240
Received: 22 Nov 2017
Accepted: 26 Mar 2018
Published online: 11 Feb 2019 *