Real-Time Client-Side Phishing Detection using Lightweight Machine Learning Models

Phishing is one of the major issues concerning online security, in which attackers attempt to gather confidential information by using fake websites with the same ”look and feel” as the actual online applications. The efficiency of the existing online phishing detection system is limited by certain limitations, including delays in database updation, failure to detect zeroday attacks, and violation of online privacy by using external servers. Keeping this issue in mind, the current paper aims to propose a novel phishing detection system called ”PhishLock.” The proposed phishing detection system is based on the use of the machine learning approach to extract lexical features based on the attributes of the URLs. The proposed system is also based on the use of the machine learning approach to extract structural features based on the Document Object Model structure of the web pages. The proposed system was developed as a prototype, and the experiments were carried out using 20,000 labeled web pages collected from PhishTank, OpenPhish, and Alexa Top Sites. The proposed system was found to have achieved 97.8.