Distributed communication model-learning architecture for anomaly detection in multi-service shared M2M area networks

This paper proposes a distributed communication model-learning architecture for detecting anomalies in multiservice shared machine-to-machine (M2M) area networks. Since M2M devices are often located at remote areas, anomaly detection is essential for M2M services to promptly react to undesirable events (e.g. a device breaks or is stolen, misused, or spoofed). To autonomously set the threshold for anomaly detection, monitoring a device's communication and learning communication model for the device are required. However, this imposes high computation burden on the network, especially when a huge number of devices are connected. Therefore, we propose a distributed learning architecture that works even in large-scale networks. Computer simulation was conducted and showed that our architecture can learn a communication model in a short time while the numbers of communications and calculations for learning remain small.