Deep learning-based detection system for cross-site scripting attacks

In modern world, many task are being transferred to virtual environments, and works which were once done manually are now being carried out through web applications. Initally, task transferred to web applications were run on local networks. However,with increasing mobility, concepts of remote work, and the need for accessibility from anywhere, applications have become accessible on the internet, making it easier for users to access while also presenting new targets for attackers. Furthermore, the concepts of local network and public network have been abandoned with the zero trust approach, and every area accessible to the application is now considered to have the same level of security. This situation has turned web applications developed without considering security concepts into open threats, assuming they will operate in secure and controlled networks. Attacks on such systems are attempted to be thwarted using signature-based IPS -IDS systems; however, attackers are constantly changing their methods to circumvent such systems. The updating speed of signature databases fall behind the development speed of new methods. The aim of this study is to develop a high-performance XSS attack detection system using deep learning systems. In the study, convolutional neural networks (CNNs), one of the new trends in the field of artificial intelligence, have been utilized. The dataset contains samples collected from OWASP and PortSwigger pages, including both malicious and benign examples. The developed system has achieved a 99% accuracy rate, demonstrating its suitability for integration into existing systems.