An Advanced Framework Leveraging Contrastive Multilevel Graph Neural Networks With Snow Avalanches Algorithm to Enhance IoT Network Security for Intrusion Detection

ABSTRACT

Internet of Things (IoT) applications are becoming increasingly popular owing to their extensive application in a variety of real‐world services. Many devices in the IoT ecosystem are connected to each other over the internet, making IoT networks vulnerable to different sorts of cyberattacks; network security and user privacy are important concerns in its deployment. In this paper, an advanced framework leveraging contrastive multilevel graph neural networks with snow avalanches algorithm to enhance IoT network security for intrusion detection (CMGNN‐SAA‐IoT‐ID) is proposed. Initially, the raw network traffic is collected from CICIoT2023 and CIC‐MalMem‐2022 datasets. The collected data are given to the preprocessing, where the Bayesian boundary trend filtering (BBTF) is applied for data cleaning, missing values, and data normalization. Then the preprocessed data are fed to feature selection using banyan tree growth optimization (BTGO) to select best features. Then the selected features are fed to intrusion detection using contrastive multilevel graph neural networks (CMGNN), which is used to detect the attacks as recon, DoS, DDoS, brute force, web‐based, spoofing, and Mirai from CICIoT2023. Additionally, the CIC‐MalMem‐2022 dataset is used to detect attacks as benign and malware. Finally, snow avalanches algorithm (SAA) is proposed to optimize the CMGNN for precisely classifying the intrusion detection. The proposed CMGNN‐SAA‐IoT‐ID is implemented, and performance metrics such as accuracy, recall, precision, computational time, and detection rate are analyzed. Finally, the performance of the CMGNN‐SAA‐IoT‐ID provides 26.68%, 25.75%, and 26.16% high accuracy and 29.08%, 30.70%, and 16.26% high precision compared to the existing models.