Probabilistic Graph Learning Based Anomaly Detection Framework for Internet of Things Security

Abstract

The blistering growth of the Internet of Things (IoT) networks has posed considerable issues of security because of the growing number of connected devices and the susceptibility of them to cyberattacks. Conventional anomaly detectors usually cannot reflect complicated communication association and dynamic behavioral trends that exist in the IoT context. This paper suggests a Probabilistic Graph Learning Based Anomaly Detection (PGL-AD) model on the CICIoT2023 data. The suggested method models IoT devices as the nodes of a probabilistic graph, with the communication relationships included as weighted edges with the probabilities of interactions. Learning through graph representation is used to learn probabilistic embeddings that incorporate structural and behavioral network traffic attributes. Probabilistic inference is carried out to calculate anomaly scores to determine abnormal communication patterns. Experimental performance shows that the proposed framework has reached $99.08 \%, 98.86 \%, 98.91 \%$, and $98.86\%$ detection accuracy, precision, recall, and F1-score respectively, and is superior to the traditional machine learning and deep learning models. The probabilistic graph learning algorithm is a good algorithm with the capability to learn network dependencies and uncertainty, to be able to detect anomalies accurately and at scale. The proposed framework is a dependable and effective measure of increasing the security of IoT and ensuring that connected devices are not affected by developing cyber threats.