Affiliations: School of Computer Science Engineering and Information Systems, Vellore Institute of Technology, Vellore, Tamil Nadu, India
Correspondence:
[*]
Corresponding author. N.C. Senthilkumar, School of Computer Science Engineering and Information Systems, Vellore Institute of Technology, Vellore, Tamil Nadu, India. E-mail: ncsenthilkumar@vit.ac.in.
Abstract: The Internet of Things (IoT) has many potential uses in the day-to-day operations of individuals, companies, and governments. It makes linking all devices to the internet a realistic possibility. Convincing IoT devices to work together to implement several real-world applications is a challenging feat. Security issues impact innovative platform applications due to the current security state in IoT-based operations. As a result, intrusion detection systems (IDSs) tailored to IoT platforms are essential for protecting against security breaches caused by the Internet of Things (IoT) that exploit its vulnerabilities. Issues with data loss, dangers, service interruption, and external hostile assaults are all part of the IoT security landscape. Designing and implementing appropriate security solutions for IoT environments is the main emphasis of this research. Within the Internet of Things (IoT) context, this research creates a Spotted Hyena Optimizer (SHO-EDLID) method for intrusion detection using ensemble deep learning. The main goal of the demonstrated SHO-EDLID method was to detect and categorize intrusions in an Internet of Things setting. It comprises many subprocesses, including feature selection, categorization, and pre-processing. The SHO-EDLID method uses a SHO-based feature selection strategy to identify the best feature subsets. It then used an ensemble of three DL models— a deep belief network (DBN), a stacked autoencoder (SAE), and a bidirectional recurrent neural network (BiRNN)— to detect and name cyberattacks. Finally, the DL models’ parameters are tuned using the Adabelief optimizer. A comprehensive simulation was run to illustrate that the offered model performed better. According to a thorough comparative analysis, the suggested method outperformed other recent approaches. Purpose of the Manuscript: To identify the best feature subsets, the SHO-EDLID method used the SHO-based feature selection method... Afterward, cyberattack identification and tracking were carried out using an ensemble of three DL models: DBN, SAE, and BiRNN. The final step in optimizing the DL models’ parameters is the Adabelief optimizer. The main comparative results: The proposed model present the Comparative analysis of SHO-EDLID algorithm with other existing systems and its outperform the performance in precision 97.50, accuracy 99.56, Recall 98.42, F-Measure.97.95.
Keywords: Security, internet of things, deep learning, ensemble learning, spotted hyena optimizer
