Machine Learning-Based Analysis of Packet Delivery Rate in LoRaWAN in Tropical Urban Environments: Case Study of Benin City, Edo State, Nigeria

Abstract

Low Power Wide Area (LPWA) communication protocols are essential for IoT due to their low power consumption and extensive communication range. LoRa technology, a key LPWA protocol, has garnered significant interest for its long-range, low-power wireless communication capabilities under the LoRaWAN standard, making it ideal for networks requiring long distances and prolonged battery life. However, few, if any, empirical studies involving the packet delivery ratio (PDR) performance of LoRaWAN networks have been carried out in the specific study location using both Machine learning and Deep learning. In our paper, the PDR performance of LoRaWAN networks in an urban tropical region, specifically Benin City, Edo State, Nigeria, was assessed using a Dragino LG02 dual-channel kit gateway with two SX1276/SX1278 radio chips, during the dry season. Experiments measured parameters such as distance, RSSI, altitude, and packet transmission data to derive PDR. Machine learning and deep learning models such as Multiple Linear Regression (MLR), Support Vector Regression (SVR), Random Forest (RF) and Artificial Neural Networks (ANN) were used to predict PDR in the test environment. It was experimented with the MLR and the ANN model that was showing the best performance after evaluation of the test set with the Root Mean Square Error (RMSE) metric, achieving error rates of 5.942 and 7.820, respectively. The path loss exponent was determined to be 3.048 for the test environment. The findings indicate successful deployment of LoRaWAN in the area, establishing key RSSI parameters and highlight that some machine learning models can serve as good predictive models for packet success rates.

Keywords: LoRaWAN, Benin City, Nigeria, PDR, Machine learning, Pathloss exponent, Tropical region