Enhancing Cloud Security and Performance: Context-Aware Group Lasso and Deep Reinforcement Learning for Attack Detection

Ensuring robust security in cloud environments is crucial due to the increasing sophistication of cyber threats.Traditional methods often fall short in balancing high detection accuracy with minimal impact on system performance.This research addresses this problem by proposing a novel Deep Reinforcement Learning with Context-Aware Group Lasso (DRL-CAGA) method for detecting and mitigating network attacks in cloud environments.The purpose of this research is to develop a system that not only accurately identifies attacks but also maintains low latency, quick response times, and efficient bandwidth usage.The novelty of the DRL-CAGA method lies in its dynamic feature selection and optimization capabilities.By integrating Context-Aware Group Lasso, the method prioritizes relevant feature groups based on their significance in attack detection, thereby reducing computational overhead.The use of Deep Reinforcement Learning allows the system to continuously adapt to evolving threats by learning optimal policies through interactions with the environment.This adaptive learning process ensures that the model remains effective even as new types of attacks emerge.The proposed method involves a comprehensive architecture that includes data preprocessing, dynamic feature selection, and reinforcement learning-based training.Evaluations are conducted using three widely recognized datasets: MAWILab, CICDDoS2019, and CTU-13.The results demonstrate that DRL-CAGA achieves superior accuracy compared to existing AI models including deep neural networks (DNNs), deep belief networks (DBNs), radial basis function networks (RBFNs), and long short-term memory (LSTM) networks, with accuracy rates of 97% for MAWILab, 95% for CICDDoS2019, and 96% for CTU-13.Additionally, the method achieves a latency of 10 ms, a response time of 12 ms, and bandwidth usage of 8 MBps, outperforming other AI-based models in these key performance metrics.