Machine Learning-Driven Phishing Detection: A Robust Browser Extension Solution

This paper addresses the evolving challenge of phishing threats with the rise of sophisticated evasion techniques. The research focuses on leveraging machine learning (ML) techniques for the automatic detection of phishing websites, providing an efficient and scalable solution to mitigate such cyber threats. This system captures the important patterns in URLs and the attributes of websites by following the technique of feature engineering, which were used to feed the machine learning models with classifications. The most important features checked were the use of suspicious domains, which leads to misleading URLs, inconsistent or unregular structure of the page, and the usage of obfuscation techniques. Models were evaluated using metrics such as F1 score and area under the receiver operating characteristic curve (AUC-ROC), showing good generalization to new data and high accuracy for detection. The study also compares the computation efficiency and detection performance of various machine learning algorithms, identifying the most efficient model for real-time phishing website detection. The work concludes by highlighting the potential of integrating these machine learning-based detection systems with web browsers and security instruments to protect end-users against real-time phishing attacks through an automated and scalable solution