Transfer Learning for Phishing Detection: Screenshot-Based Website Classification

Phishing remains a significant threat in the evolving cybersecurity landscape as phishing websites become increasingly similar to legitimate websites, complicating detection using traditional methods. This study explores AI-based solutions for screenshot-based phishing detection, utilizing the MTLP dataset and applying transfer learning with pretrained models (DenseNet, ResNet, EfficientNet, Inception, MobileNet, VGG) using the timm library. The study also discusses challenges related to phishing datasets and compares publicly available datasets, highlighting MTLP Dataset's strengths. DenseNetBlur121D was identified as the top-performing model, achieving an accuracy of 95.28%, a recall of 95.38%, a precision of 93.42%, and an F1 score of 94.39% when applied to the entire MTLP dataset. Both the model code and dataset are publicly available, providing a valuable resource for further research and development in this domain.