Prediction of Phishing Websites Using Machine Learning Techniques

A phishing attack is a cyber attack that generates a fake website acting as a trusted website to steal sensitive information like passwords and credit card information with the motive to misuse it. Phishing attacks have increased over time and differentiating a phishing site from a legitimate one is becoming increasingly difficult. Therefore, checking the legitimacy of visited web pages is a crucial task to secure customers' identities and prevent phishing attacks. The literature is rich with machine learning-based techniques for detecting phishing websites. However, all the present methods have shortcomings as they tend to focus on a particular feature set, thereby prone to misclassification when implemented in real-time. Hence, in this work, we propose a novel 3-level pipeline architecture to detect phishing websites with significant performance using machine learning techniques. The proposed architecture combines three methods of detection namely domain-based, content-based, and visual similarity-based. The domain-based method classifies websites based on domain details like URL, IP Address, Google Indexing, etc, whereas, the content-based model uses the content text for classification. The visual similarity model compares the screenshot of the phishing website with the closest legitimate website. XGBoost, Logistic Regression, and Triplet Network gave the best performance for the domain-based model, the content-based model, and the visual similarity model, respectively. The proposed three-layer architecture is implemented along with a Google Chrome extension to enable real-time detection. The proposed model achieved 96. 429% accuracy and 98. 989% recall on the considered real-time dataset and performed better compared to all existing single-layer architecture-based methods.