An Assessment System for ML-Based XSS Attack Detection Models Between Accuracy Coverage and Data

The huge capability of ML-based solutions has impacted various areas including cybersecurity. Those solutions can detect and respond to several threats by offering advantages over traditional rule-based methods. A serious web security threat is Cross-Site Scripting (XSS) attack, it occurs when an attacker inserts malicious content into a vulnerable web application to perform unauthorized actions. Out of the various XSS attack detection techniques and methods, ML-based models have a profound ability to achieve impressive results and improved accuracy. However, an evaluation and validation of those approaches is required. In this study, we present an assessment system for pre-existing ML-based XSS attack detection models. Our assessment system contains two phases, the first phase aims to understand the model’s architecture, and the second phase ensures the model’s efficiency across unseen datasets. In experiments, we applied our evaluation system to an existing model, results highlight some limitations such as integrity, interpretability, overfitting, and false positive rates.