Unveiling XSS Threats: A Bipartite Graph Approach with Ensemble Deep Learning for Enhanced Detection

Abstract

Cross-Site Scripting (XSS) attacks are a common source of vulnerability for web applications, necessitating scalable mechanisms for detection. In this work, a new method based on bipartite graph-based feature extraction and an ensemble learning classifier containing CNN, LSTM, and GRU is introduced. Our proposed bipartite graph model is novel as the payloads constitute the first set, while the words constructing the payloads comprise the second set. This representation allows structural and contextual dependencies to be extracted so the model can recognize complex and obfuscated XSS payloads. Our method surpasses state-of-the-art methods by having 99.97% detection accuracy. It has a significantly increased ability to detect complicated payload variations by utilizing co-occurrence patterns and interdependence between smaller payload parts through the adoption of these bipartite features. In addition to improving the F1-score, recall, and precision associated with such methods, it also demonstrates the adaptability of graph-based representation in cybersecurity applications. Our findings highlight the possibility of integrating ensemble classifiers and refined feature engineering into a scalable, precise XSS detection system.