A novel optimal deep learning approach for designing intrusion detection system in wireless sensor networks

A wireless sensor network contains many nodes to collect and transfer data to a primary location. However, wireless sensor networks have several security issues because of their resource-constrained nodes, deployment tactics, and communication channels. As a result, detecting intrusions is crucial for strengthening the safety of wireless sensor networks. Naturally, any communication network will need the services provided by a network intrusion detection system. Despite their everyday use in intrusion detection systems, the efficacy of machine learning (ML) approaches needs to be improved for handling asymmetrical attacks. This article proposes an intrusion detection system based on an Improved deep neural network (IDNN) to solve this issue and enhance performance. Using the global search strategy of the coyote optimization algorithm (COA-GS) on the KDDCup 99 and WSN-DS datasets, the following hyperparameter selection techniques are used to determine network topologies and the optimal network parameters for DNNs. The most efficient algorithm for detecting future cyberattacks can be chosen by conducting such research. Extensive studies comparing COA-GS-IDNNs and other standard machine learning classifications on a large number of openly accessible malware benchmark datasets are presented. Extensive experimental testing demonstrates that DNNs outperform conventional machine learning classifiers at real-time monitoring network activity and host-level events to detect and prevent intrusions.The experimental outcomes demonstrate that the suggested COA-GS-IDNN model increases the accuracy ratio of 95 %, the precision ratio of 94 %, recall ratio of 96 %, F1-score ratio of 95 %, ROC AUC ratio 98 %, detection time of 1.0068754, and delay of 0.8016 ms compared to other existing models.