Advanced Ensemble Deep Learning Model Integrated with Metaheuristic Optimisation for Secure and Reliable Intrusion Detection in Wireless Sensor Networks

With the growth of reliance on the internet and the transfer of most businesses to present remote services, the problems in protecting the network and identifying attacks rapidly become more prominent, as the attack surface and cyberattacks improve in response. The current Wireless Sensor Networks (WSNs) intrusion detection methods that utilize Machine Learning (ML) techniques to detect previously known attacks use single layers of recognition, which means an expensive algorithm must be performed before identifying any suspicious action. Network Intrusion Detection Systems (IDS) present a type of service to the system, and it becomes unavoidable for some communication systems. ML methods are extensively applied in IDS; still, the performance of ML methods is less adequate while processing unbalanced attacks. This paper presents an Advanced Ensemble Deep Learning Model Integrated with Metaheuristic Optimization for Secure and Reliable Intrusion Detection (AEDL-MORID) methodology. The main objective of the AEDL-MORID methodology presents strong potential for real-time deployment in resource-constrained WSN environments, strengthening network resilience against sophisticated cyber threats. The AEDL-MORID method starts with data pre-processing techniques involving the handling of missing values and min-max normalization to ensure clean and consistent input for the learning models. For dimensionality reduction, the dung beetle optimization (DBO) method is utilized to detect the most informative features effectively. In addition, an ensemble classification method integrating Bidirectional Long Short-Term Memory (BiLSTM), Graph Convolutional Network (GCN), and Stacked Denoising Autoencoder (SDAE) is employed for attack detection. To further improve ensemble classification performance, the model parameters are fine-tuned using the Improved Crow Search Algorithm (ICSA) method. The experimentation of the AEDL-MORID model is conducted on the WSN-DS dataset. The experimental validation of the AEDL-MORID system indicated a better accuracy of 99.81% compared to recent techniques.