Copyright © 2026 Authors retain the copyright of this article. This article is an open access article distributed under the Creative Commons Attribution License which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.

@article{207582,
        author = {Afreena Sathaf M and Kavi Ranjani and Aashika},
        title = {AN INTELLIGENT PHISHING WEBSITE DETECTION SYSTEM USING CNN WITH MULTI-HEAD SELF ATTENTION},
        journal = {International Journal of Innovative Research in Technology},
        year = {2026},
        volume = {13},
        number = {3},
        pages = {1743-1750},
        issn = {2349-6002},
        url = {https://ijirt.org/article?manuscript=207582},
        abstract = {Phishing websites are an ever-changing and prevalent risk to the area of the cybersecurity system as they allow attackers to collect sensitive user credentials, financial information, and personal data by spoofing the web interface. Detecting using conventional methods, including blacklist-based and rule-based methods, has serve weakness against those related to zero-day and new phishing campaigns. Moreover, the imbalance of classes in the existing datasets remains to lower the performance of the current machine learning and deep learning solutions. In this paper, an intelligent phishing site detection framework has been introduced, employing a Convolutional Neural Network (CNN) and a Multi-Head Self-Attention (MHSA) system, as a set of features to be extracted and patterns to be identified are based on URLs. The CNN part itself self-trains on discriminative features using the raw sequences of URL characters, and the self-attention components selectively enhances the discriminative features of URL patterns in order to achieve better classification accuracy and low false positive probability. In order to alleviate the negative side effects of the class imbalance, a Generative Adversarial Network (GAN) is used to generated realistic samples of phishing URLs to augment the training corpus and improve generalization of the model. Large-scale experiments reveal that the suggested system is more that the suggested system is more accurate, precise, recalls and F1-score as compared to baseline strategies, as well as real-time and zero-day phishing is demonstrated. The framework provides a scalable and adaptive framework that is appropriate to integrate into dynamic adversarial settings.},
        keywords = {detection phishing, convolutional neural network, multi-head self-attention, generative adversarial network, deep learning, and cybersecurity.},
        month = {August},
        }