Deep Q-Learning based approach for DoS attacks detection

Abstract

With the continuous advancement of attack detection techniques, Denial of Service (DoS) attacks have become increasingly diverse and frequent, causing network disruptions and threatening their availability and reliability. This highlights the need for effective defensive strategies to reduce financial losses and protect digital service infrastructures. In this context, a new detection system is proposed based on Reinforcement Learning techniques. This system enables modeling of the network security environment to detect and counter DoS attacks. The agent in reinforcement learning acquires policies that allow it to make a sequence of decisions to progressively improve its capabilities and efficiency, through continuous interaction with a simulated network environment. The proposed system works to accurately distinguish between correct and incorrect classifications, using an effective classification mechanism that enhances result accuracy and reduces error rates. The system also applies a methodology to extract distinctive data features and present them for processing, thereby enhancing the ability to dynamically and effectively adjust defensive behavior. Results show that the proposed system is capable of recognizing evolving attack patterns, contributing to the development of a more flexible and proactive network defense against DoS threats.