Automated Long‐Short‐Term Echo State Self‐Attention Network Model for Clone Node Attack Detection in Edge‐Assisted Wireless Sensor Networks

Wireless sensor networks (WSNs) face several security threats, including clone node and cyberattacks. These threats should be detected and mitigated to maintain the confidentiality, integrity, and dependability of data transfer within the network. In this work, a new automated deep learning model is proposed for clone node and cyberattack detection using channel state information (CSI) and darknet traffic features, respectively. Initially, channel responses among sensor nodes are investigated as a unique spatial and temporal pattern. Then, the labels of offline training samples for clone nodes are generated by analyzing the differences in channel characteristics using a threshold‐based method. Then, the channel characteristics of the clone nodes will be integrated with darknet traffic features to monitor the networks for other cyberattacks. This research introduces a novel long‐short‐term echo state self‐attention network (LSTECS‐SA) to detect the clone node and other unknown cyberattacks. This model uses multi‐scale echo‐state networks (ECN) for capturing the multi‐scale temporal properties of traffic features and a self‐attention network for assigning feature weightage. Also, the hyperparameters of the LSTECS‐SA are optimized using the light spectrum optimization (LSO) algorithm, which can minimize the error rate of the attack detection model. According to the simulation results, our technique can detect clone nodes with a great accuracy rate when the threshold is set appropriately. Additionally, the suggested LSTECS‐SA outperforms the existing models by achieving an accuracy of 97.56% on the generated dataset.