A novel intrusion detection system: integrating greedy sand cat swarm optimization and dual attention graph convolutional networks

The rise of smart devices and network vulnerabilities has led to a surge in cyber-attacks. Detecting and classifying malicious traffic is vital for system security. This paper proposes a novel framework for intrusion detection using advanced machine learning techniques to improve cybersecurity. The framework initiates comprehensive data collection from the BoT-IoT and NSL-KDD datasets, followed by rigorous data pre-processing steps including normalization, label encoding, and outlier identification. Feature extraction is performed to capture key characteristics of the data, and dimensionality minimization techniques are applied to improve computational efficiency. A feature selection process is executed using the Greedy Sand Cat Swarm Optimization algorithm that identifies the most informative features for analysis. These features are then processed by a Dual Attention Graph Convolutional Neural Network, designed to reveal complex patterns in network traffic data. The framework outperforms traditional methods with the accuracy and precision of 99.19% 99.08% respectively. Overall, these findings highlight that the classification performance of the proposed model is highly accurate, making a significant contribution to improving intrusion detection and network security.