Greedy Behavior Detection With Machine Learning for LoRaWAN Network

LoRaWAN (Long Range Wide Area Network) has garnered significant attention within the Internet of Things (IoT) due to its ability to establish a wireless network with massive devices over long distances while minimizing energy consumption. Our previous work shows its suitability for Intelligent Transportation Systems (ITS) scenarios. However, the utilization of the Aloha MAC protocol presents a challenge for LoRaWAN as it grapples with the presence of compromised nodes. These nodes may engage in greedy behaviors, disregarding network regulations to enhance their own performance or acquire additional network resources, and are often difficult to detect. This research contributes to machine learning-based greedy behavior detection methods. After proposing several end-to-end (E2E) methods with different ML algorithms, EDLoG (Encoder-based detection method of LoRaWAN Greedy behaviors) is proposed. It is a greedy behavior detection method combining a Multilayer Perceptron (MLP) encoder network and a statistical abnormal detection algorithm. The performance evaluations are conducted using simulation data under different scenarios given by MELoNS, a Modular and Extendable Simulator for the LoRaWAN Network developed in our previous work. The results show that the proposed method gives a detection recall 15%-20% higher than the baseline method by keeping a high detection precision. Moreover, the proposed methods show high timing efficiency with a running time much smaller than LoRaWAN's time scale, making the method easily deployed to a real LoRaWAN network.