Enhancing web attack detection efficiency based on natural language processing techniques

Abstract

Web security continues to pose significant challenges for organizations and individuals worldwide. The rapid evolution of technology has made website protection increasingly complex against sophisticated cyber attacks. Delayed vulnerability remediation often results in severe consequences, including authentication bypass, data breaches, information theft, and complete system compromise. Despite extensive research and proposed attack prevention and mitigation solutions, existing approaches demonstrate limitations due to high false-positive rates. Recent advances in artificial intelligence have shown promising results in improving detection capabilities, especially for zero-day attacks. In this study, we propose the Lightweight Deep Web Learning (LDWL) model based on Natural Language Processing (NLP) for feature extraction and employ multilayer neural networks for classification. We evaluated the model on the HTTP Param dataset and achieved excellent results, with all metrics reaching 99.99% accuracy