Efficient abnormal behavior detection in information-centric internet of things using SVM

Information-centric Internet of Things networks play a vital role in modern systems due to the increasing number of connected devices and the massive volume of data generated. However, due to complex node interactions and diverse applications, these networks are prone to abnormal traffic behaviors that can jeopardize network performance and security. Despite numerous efforts to analyze traffic and identify normal and abnormal behaviors, existing methods often lack sufficient accuracy and experience performance degradation when faced with sudden traffic changes. Furthermore, many approaches require heavy computation, which is incompatible with the resource constraints of IoT nodes. Efficient behavior prediction in information-centric IoT networks is critical for optimizing energy use and maintaining network stability. This paper proposes a Support Vector Machine (SVM)-based model that classifies node behavior as normal or abnormal using features like packet reception rate, delay, and energy consumption. The model is evaluated on the MDC dataset (10,000 traffic samples from 500 nodes) and the public UNSW-NB15 dataset. Compared to Support Vector Regression (SVR) and SVM with denoising, the proposed model improves accuracy by 12% and 9%, respectively. It achieves a 15% higher F1-score and maintains stable performance under increasing delay and packet load. Statistical tests (Wilcoxon, Friedman) confirm the significance of performance differences.