Leveraging Metaheuristic Optimization-based Significant Feature Selection with Hybrid Deep Learning Approach for Intrusion Detection in Wireless Sensor Networks

Wireless sensor networks (WSN) have numerous latent applications. Besides, in multiple states, WSNs gained more attention from opponents. They became vulnerable to few threats since they were utilized in open and insecure settings and are established low-cost small devices. Defensive mechanisms are employed to defend WSNs against a few kinds of threats. However, there are a few threats and no recognized prevention methods. For this case, utilizing a few devices for intrusion detection is vital. Also, to prevent the intruder from harming the system, the intrusion detection system (IDS) can get data related to the attack model, aiding in the growth of prevention models. IDS are critical tools in preserving WSNs against these trials. Machine learning (ML) and deep learning (DL) models are getting high attention while executing an IDS for WSNs. This study proposes a Feature Selection-Based Metaheuristic Optimization Algorithm for Intrusion Detection in Wireless Sensor Networks (FSBMOA-IDWSN) model. The presented FSBMOA-IDWSN model is a highly efficient and adaptive cybersecurity solution for securing WSNs against evolving cyber threats. The z-score normalization is initially utilized for data pre-processing to standardize feature distributions and improve model stability. Furthermore, the dynamic arithmetic optimization algorithm for feature selection (DAOAFS) method is applied to select the most related features. The convolutional long short-term memory with autoencoder (ConvLSTM-AE) technique is employed for intrusion detection and attack classification. Finally, the hyperparameter fine-tuning of the ConvLSTM-AE model is accomplished by implementing the salp swarm optimization (SSO) technique to achieve superior accuracy. A wide range of experiment analyses was performed, and the results were examined under the WSN-DS dataset. The comparison study of the FSBMOA-IDWSN technique portrayed a superior accuracy value of 99.67% compared to existing approaches.