Attack detection and network recovery using Correlation Aware Lotus Effect Hierarchical Dual Graph Neural Networks

Abstract

The recent advancements in Software Defined-Wireless Sensor Networks (SD-WSNs) have revealed the susceptibility of the systems to future enhancement in cyber–physical attacks. However, intrusion detection creates an anomalous behavior that bypasses traditional detection systems. To address these problems, the Correlation Lotus Effect Hierarchical Dual Graph Neural Networks (CLEHDGNN) is a new proposed method to guarantee the correct classification of various types of attacks. This approach utilizes a correlation-sensitive framework is used to efficiently discern the relationships between features for attack detection. Density-Based Spatial Clustering (DBSC) is employed for clustering time-oriented data, whereas the selection of cluster heads is achieved through a Particle-Guided Metaheuristic Algorithm (PGMA) to ensure effective cluster management. This method serves to facilitate and enhance the rapid identification of emerging attack patterns for improved detection accuracy performance. This approach is based on the integration of the Lotus Effect Optimization Algorithm, which improves the identification and classification of attacks. The proposed method has shown astonishing results, reaching a maximum accuracy of 98.5%, F1-score of 97.9%, precision of 98.6%, recall of 98.95%, and specificity of 99.6 as compared to existing techniques. Overall, the proposed model holds significant potential for bolstering the reliability of security in cyber–physical systems.