Enhancing phishing site detection and prevention through machine learning with image classification
2025, vol.17 , no.4, pp. 87-98
Article [2025-04-09]
Distinguishing expertly designed phishing sites from genuine ones is difficult, likely contributing to the continued victimization of users by these attacks. The challenge of creating a universal algorithm to detect these malicious pages has resulted in the development of various tools to protect users. This paper introduces an innovative method for detecting and preventing phishing websites using machine learning, which is implemented in a newly developed browser plugin called NotPwned. Additionally, tools have been created for data collection and training computer vision models. A classifier has been designed to identify login forms based on submitted screenshots. The most well-known algorithms for object detection in images have been analysed and compared. Furthermore, a model has been trained to recognize different logos. The research concludes with a comparison of the system's capabilities against those of competing solutions, demonstrating the effectiveness of the proposed method and the resulting product.
anti-phishing, computer vision, object detection, plug-in, phishing
https://doi.org/10.59035/SABE2922
Kaloyan Manev, Milen Petrov, Adelina Aleksieva-Petrova. Enhancing phishing site detection and prevention through machine learning with image classification. International Journal on Information Technologies and Security, vol.17 , no.4, 2025, pp. 87-98. https://doi.org/10.59035/SABE2922