Multi-Feature Extraction-Based Phishing Website Detection Using Ensemble Learning

Phishing attacks, in which users are duped into engaging with websites that seem authentic, are a common method of cyber attacks. These are pages designed to successfully fool a person by appearing real. Phishing attacks continue to be a serious cybersecurity risk, resulting in significant losses of personal and financial information. Therefore, it’s critical to identify a solution that can automatically reduce such security risks. To address the limitations of current detection methods that often rely on limited feature sets, we propose a comprehensive approach that integrates five types of URL features including NLP-based, address-based, abnormal, domain-specific, and HTML-based. Further, Recursive Feature Elimination with Cross-Validation (RFECV) is employed for optimal feature selection to assist models in concentrating on the dataset’s most pertinent information. Different machine learning as well as deep learning models are implemented using the prepared dataset. Our investigation indicates that the proposed ensemble model, combining Random Forest, XGBoost, and CatBoost, achieved the highest accuracy of 88.90% on a unified dataset and outperformed prior studies on three separate datasets. This integrated approach significantly will enhance phishing detection accuracy, providing a robust cybersecurity solution. Future research will explore additional features to improve detection capabilities further.