Machine Learning‐Based Malicious Web Page Detection Using Generative AI

The accelerated growth of the Internet has changed the basic ways of accessing information, communicating, shopping, and doing business by individuals and organizations. Web services have become a part of everyday life, as they help in online banking, remote work, and online shopping. However, this online convenience has also brought about bad players in the ministry. Dubious web pages are also an increasing way in which cybercriminals gain access to the system to commit phishing attacks, infect files with malware, and steal valuable information. Such pages appear as replicas of real websites and can be difficult to identify using conventional security measures. Traditional detection mechanisms, such as signature-or heuristic-based detectors, are not very effective in keeping up with the advanced and dynamic methods used by attackers. To this end, machine learning (ML) and generative AI (GenAI) seem to be effective substitutes for traditional tools. Large datasets can be learned by ML models to identify other minor trends and outliers that can be used to identify malicious actions. Meanwhile, GenAI has the potential to create realistic phishing content that can be used to robustly train and stress-test detection systems. This chapter investigates the hybridization of these technologies by incorporating them into a hybrid detection framework that performs better than traditional approaches. In this lesson, one learns about the merits of ML classifiers, the use of GenAI as a generator of adversarial content, and how the two can improve detection rates and flexibility. Moreover, this chapter mentions the existing gaps and provides future research directions so that potential cyber threats can be dealt with and curbed successfully. We have made all the suggested changes in the Results section, made the dataset division clearer, analyzed the types of works that exist in different languages, explained why we chose the embedding models, clarified the purposes of the explanation models, and spelled out our contributions more clearly. Furthermore, we emphasize how our work differs from previous studies to demonstrate its originality.