Feature selection approach to detect phishing website using machine learning algorithm
Siti Nur Aqilah Kamarudin, Isredza Rahmi A. Hamid, Cik Feresa Mohd Foozy, Zubaile Abdullah · AIP conference proceedings · 2022
Phishing is a form of scam and social engineering attack that often used to steal user information including login credentials, social security number, and credit card number. For the past few years, the most common phishing activities were conducted using email phishing, Short Message Service (SMS) phishing, phone call phishing, and website phishing. According to Google's Transparency Report, an average of 46,000 new phishing websites are detected every week in 2020. To overcome the issues, machine learning algorithm is used to show the differences between the legitimate link and the phishing link. In this paper, we consider hyperlink-based approach to detect phishing website. We focused on hyperlink as it is the main contributor in phishing related to email, SMS, and website phishing. These hyperlink features selection approach used Correlation, Gain Ratio, Information Gain, OneR, ReliefR, and Symmetrical Uncertainty algorithm. The features selected are tested on machine learning algorithm using WEKA tools. The performance of features selection approach is shown in terms of accuracy, precision, true positive, and false positive. As a result, Information Gain is the best classification algorithm using top 5 features and top 20 features tested on Random Forest algorithm with 94 percent and 95.6 percent respectively. For 10 features ranked using ReliefR algorithm recorded its best tested on Random Forest algorithm with 94.6 percent.