WebGuardRL: An Innovative Reinforcement Learning-based Approach for Advanced Web Attack Detection

Web-based applications are often potential targets for attackers due to the important data and assets that they manage. With the explosion and increasing complexity of recent attacks aiming at these applications, traditional security solutions such as intrusion detection systems (IDS) or web application firewalls (WAF) become ineffective against unpredictable threats. Meanwhile, in the trend of applying AI techniques to achieve practical effectiveness in various fields, cutting-edge reinforcement learning (RL) has also gained more attention for its promising applications, one of which is sophisticated attack detection. In this study, we introduce an RL-based model, named WebGuardRL, to detect multiple advanced web attacks by analyzing URLs in HTTP requests containing various attack types. To achieve this, our model is equipped with the capability of representing URLs that differ from attack to attack in the same form for use in RL training. The experimental results and comparisons with other methods indicate the high accuracy and remarkable capability of our WebGuardRL in web attack detection.