Enhancing Phishing Universal Resource Locator Detection Systems Using Hybrid Machine Learning Classifiers and Data Balancing Techniques

Phishing attacks have become a significant threat to online security, with cybercriminals using sophisticated tactics to evade detection. Traditional phishing detection systems often relied on rule-based approaches, which can be limited in their effectiveness. This research uses machine learning (ML) to detect phishing universal resource locator (URL) by evaluating the performance of advanced classifier for URL phishing detection using both original and balanced datasets. The classifier assessed is Random Forest. In conjunction with data balancing techniques like SMOTE and Resample, on the original dataset, it achieved an accuracy of 98.10% with a Kappa statistic of 96.20%, requiring 1.93 seconds for training and 0.05 seconds for testing. For the balanced dataset, the performance slightly improved, achieving an accuracy of 98.20% and a Kappa statistic of 96.40%, with reduced training time of 1.30 seconds and testing time of 0.03 seconds. Hence, balancing original dataset can significantly improve URL phishing detections using ML.

© 2024 IEEE. Personal use of this material is permitted. Permission from IEEE must be obtained for all other uses, in any current or future media, including reprinting/republishing this material for advertising or promotional purposes, creating new collective works, for resale or redistribution to servers or lists, or reuse of any copyrighted component of this work in other works.