Enhanced Phishing URL Detection Using Leveraging BERT with Additional URL Feature Extraction

Threats, including phishing, are still a widespread and growing problem in today's digital society because fraudsters use increasingly complex techniques to deceive users and gain illegal access to sensitive information, effective phishing detection systems are essential for protecting individuals and enterprises. In this study, we provide a novel method for phishing detection that takes advantage of the capabilities of the BERT model and URL feature extraction. In testing employing a dataset of 200,000 URLs, including original and phishing URLs, our technique achieved a remarkable accuracy of 97.32 percent. By properly preparing the information and using the BERT tokenizer to tokenize the URLs, the contextual understanding was made possible. Additionally, the model's ability to discern between legitimate URLs and phishing efforts was enhanced by the capturing of structural features using URL feature extraction techniques. The high degree of accuracy attained by our technique shows that it is useful in phishing attack detection and prevention, offering a strong defense against fraudulent online activity and boosting user security.