Scalable Machine Learning Approach for Realtime Detection of Xss Attacks in Web Applications

Cross-Site Scripting (XSS) is a pervasive web application vulnerability that allows attackers to inject malicious scripts into web pages, compromising the security of other users. Designing a cloud-based XSS detection system that effectively handles large-scale web applications poses challenges in both accuracy and operational efficiency. Most existing solutions primarily focus on detection accuracy, often overlooking critical aspects such as visualization and attack detection efficiency for high volumes of transactions for multiple web applications. In this paper, we present a SMARTX model providing a solution for detecting XSS attacks using a machine learning-based approach, leveraging a Multi-Layer Perceptron (MLP) classifier with optimal accuracy-related parameter selection. To enhance detection speed and scalability, we employ multiprocessing techniques for real-time threat analysis. Additionally, we introduce a graph-based model, implemented using the Neo4j database, to efficiently store and visualize detection results across multiple cloud-based web applications. This model establishes correlations between hosts, clients, and detection outcomes, enabling actionable insights for administrators. Our architecture integrates both client-side and server-side detection mechanisms, providing robust protection against various XSS injection vectors. Experimental results demonstrate that SMARTX model achieves high detection accuracy while optimizing processing times, making it highly suitable for real-world deployment in dynamic web environments.