PhishFind: A Machine Learning-Based System for Real-Time Phishing Detection

Phishing attacks continue to exploit user trust, posing significant risks to individuals and organizations through increasingly sophisticated tactics. Existing detection tools often lack real-time analysis or transparent explanations, leaving a gap in effective browser-based protection. This work introduces PhishFind, a browser widget designed to address these limitations by integrating advanced machine learning and explainable AI. As part of this contribution, we developed PhishingLong, a continuously updated dataset of phishing and legitimate websites. Leveraging this dataset, the system applies a Gradient Boosting classifier to analyze URLs and web content, while a semantic module based on the ChatGPT API provides users with clear, human-readable explanations for suspicious sites. In evaluation, Gradient Boosting achieved an F1-score of 97.34%, and user testing demonstrated high acceptance, particularly for usability and alert clarity. Overall, PhishFind demonstrates the potential of combining robust detection with explainable feedback to enhance user protection against phishing in real time.