Ensemble Learning Framework for Phishing URL Detection and Leveraging LLMs for Explainability

Recent trends show that Phishing has become a main method for attack by which a malicious attacker can steal sensitive personal information, such as bank card numbers, by imitating people or respected institutions like banks, email providers, social networks, and other web services. Being one of the oldest types of online threats, phishing attacks remain a significant problem regarding online security. Despite the existence of several detection systems, phishing as an attack is difficult to completely resolve due to the ever-evolving nature of these attacks. This paper proposes a stacking based multilayered ensemble learning model which is trained for URL-based phishing detection. This model incorporates both boosting algorithms as well as deep learning techniques, in order to improve classification accuracy using a stacking approach. The predictions from the models in the first-layer are sent as the parameters for higher-level models, refining our resultant predictions through iterative learning. Results demonstrate that the resulting model can detect phishing websites, outperforming existing models, with recall, accuracy, recall, and F1-score exceeding 99.9%, while using fewer parameters making it applicable in a real time environment.