A Review of Phishing URL Detection Using Machine Learning Classifiers

Abstract

Phishing is a rapidly increasing threat to the modern internet where phisher mimics a legit web-page to get user into the phishers cage. The aim of phisher is to get sensitive records or information about the user such as credit card details, email and passwords. Anti-Phishing Working Group recently reported that 86,276 unique phishing URLs have been detected in the month of September, 2019. Thus, to resolve this threat, different methods and techniques have been proposed such as blacklist, whitelist, heuristics of URL, content based, image processing as well as Machine Learning based. Machine Learning (ML) is a modern technique and its algorithms have better efficiency, accuracy and performance. Thus, this paper critically reviews and evaluates Machine Learning based classifiers on the basis of datasets used, feature extraction techniques and performance measures used for the detection of phishing URLs. Moreover, the literature review reveals that every ML base approach has its advantages and limitations; hence, suggesting one classifier over other is challenging. However, Random Forest (RF) and Support Vector Machine (SVM) are mainly used classifiers in the literature of this study, where RF achieved the highest accuracy using larger dataset. In contrast, SVM achieved the second highest accuracy using smaller dataset. This research concludes that RF is an efficient approach for the detection of phishing URLs. Finally, insights are provided at the critical evaluation section and this research proposed extracted features that can enhance the literature in order to develop an effective URL phishing detection method.