Multiclass Classifications of Web Injection Attack on Http Request Data Using Machine Learning

Web applications are constantly facing attacks which try to inject malicious code into vulnerable systems. While existing security measures provide some protection, they often struggle to keep pace with ever-growing attack techniques. This paper proposes a novel approach to web application defense that leverages machine learning models, specifically Random Forest and Convolutional Neural Networks (CNNs), to analyze HTTP request data and accurately identify various types of attacks, including SQL injection and cross-site scripting (XSS). By employing a comprehensive dataset with modified payloads, our experimental results demonstrate significant improvements over traditional rule-based methods. Our system achieves an impressive overall accuracy of 0.996, outperforming both Random Forest and ModSecurity, particularly in classifying normal, SQL injection, and XSS requests. This advancement offers a significant contribution to web application security by providing a more comprehensive and adaptive defense against evolving attack strategies.