ADVERSARIALLY ROBUST AND REAL-TIME EXPLAINABLE DETECTION OF CROSS-SITE SCRIPTING ATTACKS By USING ADAPTIVE MACHINE LEARNING

Abstract

Cross-site Scripting (XSS) is considered one of the most popular and constantly developing vulnerabilities of the modern web-based applications, posing serious threats to the modern web-based infrastructures. In spite of the recent research, performed on hybrid and context-based machine learning that design approaches to detect XSS, most of the available can be applied in offline environments, has some of a lack of resilience to adversarial payload obfuscation, and has low interpretability. These weaknesses reduce their effectiveness in the real time deployment scenarios whereby the enemies continuously evolve their strategies. This paper introduces explainable, real-time, and adaptive Adversarially robust framework of XSS attack detection. In the suggested system, adversarial payload generation is integrated with online incremental learning and instance-level explainable artificial intelligence that will increase detection resilience and the level of operation transparency. Online learning model based on the context-sensitive characteristics that are obtained during the analysis of URLs, HTML structuring, and JavaScript execution behavior can be continually updated with no need to completely retrain the model. The explainable AI methods are necessary to provide instance-level explanations and enable security analysts to understand individual detection decisions in real-time in order to ensure trust and usability. Through experimental assessment, it is proven that the suggested framework experiences a high detection rate during adversarial obfuscation and that it significantly decreases the inference latency in comparison to offline models. These results support the validity of implementing adaptive and explainable distributed systems of XSS in detecting systems in dynamic web applications. This study advances the state of the art, in the field of intelligent web security defense by concurrently building up the adversarial robustness, real-time performance, and interpretability.