Phishing Website Detection: A Systematic Review of URL-Based and Deep Learning Approaches

Abstract

Phishing websites have reached to be one of the most important cybersecurity threats in the digital ecosystem. They achieve this through impersonating reputed online services and through this technique, they get hold of the username, password, bank details, and even personal identification data of an unsuspecting user. The successful phishing attacks might lead to financial destruction, losing one's identity, privacy invasion, and the organization suffering from a bad reputation. The application of traditional phishing detection methods such as blacklists and rule-based heuristics has not been very effective, especially with the newly-created and fast- changing phishing websites. The development of deep learning (DL) has helped to a great extent in the improvement of the phishing detection systems because it has allowed for automated feature learning, better generalization, and real-time classification. The discussion on URL feature–based phishing detection is gaining traction and is being considered due to its minimal processing requirements, the possibility of being available at an early stage and thus being suitable for real-time deployment. Unlike the content-based or behavior-based detection methods, the URL-based method does not necessitate webpage rendering or external queries thus making it very efficient for browser and network-level security. This paper reviews and presents a detailed systematic analysis of deep learning–based real-time phishing website detection systems that utilize URL features. Its focus is on the evolution of phishing detection methods, URL feature representations, deep learning architectures, as well as system-level considerations. Furthermore, the paper discusses key challenges such as zero-day attacks, adversarial manipulation, concept drift, and model interpretability, while highlighting promising future research directions. By consolidating existing research into a structured review, this work aims to serve as a comprehensive reference for researchers and practitioners working in intelligent cybersecurity systems.