IGSA-SAC: A Novel Approach for Intrusion Detection Using Improved Gravitational Search Algorithm and Soft Actor-Critic

Lizhong Jin ^* Rulong Fan Xiaoling Han Xueying Cui

• Taiyuan University of Science and Technology, Taiyuan, China

Background: Network intrusion detection is a critical component of maintaining network security, especially as cyber threats become increasingly sophisticated. While deep learningbased intrusion detection algorithms have shown promise, they often struggle with highdimensional datasets containing outliers, anomalies, or rare events. This study addresses these challenges by proposing a novel approach that combines the Improved Gravitational Search Algorithm (IGSA) with the Soft Actor-Critic (SAC) reinforcement learning algorithm, aiming to enhance detection accuracy and computational efficiency. Methods: We introduce the IGSA-SAC intrusion detection model, which leverages an enhanced Gravitational Search Algorithm (IGSA) to improve robustness against outliers and dynamically adjust the explorationexploitation balance. This is achieved through fitness normalization with an Adaptive Search Radius and a sigmoid function to modulate the gravitational constant. The IGSA-SAC method effectively navigates the search space to identify the most relevant features for intrusion detection, reducing dimensionality and computational complexity. Additionally, we design a reinforcement learning reward function to guide the learning process, encouraging the agent to improve detection effectiveness while minimizing false alarms and missed detections. Results: Experiments were conducted on the NSL-KDD and AWID datasets to evaluate the performance of IGSA-SAC. The results demonstrate that IGSA-SAC achieves an accuracy of 84.15% and an F1-score of 84.85% on the NSL-KDD dataset. On the AWID dataset, IGSA-SAC surpasses 98.9% in both accuracy and F1-score, outperforming existing intrusion detection algorithms. Conclusions: The proposed IGSA-SAC method significantly improves intrusion detection performance by effectively handling high-dimensional datasets and reducing computational complexity. The results highlight the potential of IGSA-SAC as a robust and efficient solution for real-world network intrusion detection systems, offering enhanced accuracy and reliability in identifying cyber threats.