Find my phish: phishing detection browser extension using deep learning methods by merging datasets

Phishing remains one of the most important cybersecurity threats, with millions of phishing websites emerging every month. Despite years of research and advancements in detection techniques, phishing attacks continue to evolve, targeting various sectors such as business and healthcare. This study explores phishing detection using machine learning and deep learning by focusing on raw URL structures rather than email-based attacks. We analyze and merge three publicly available datasets to extract and evaluate key features that contribute to effective phishing detection. The datasets include URL-based phishing indicators, structural characteristics, and binary classification labels. After feature extraction and selection, we identify 32 features, such as URL length, number of special characters, and domain tokens, that significantly impact model performance. Our feature importance analysis highlights URL length as the most influential factor in differentiating phishing from legitimate websites. Using deep learning models trained on the newly merged dataset, this study aims to improve the accuracy and adaptability of phishing detection in various real-world scenarios. The findings emphasize the need for robust, automated feature extraction techniques to enhance phishing prevention efforts.