Hybrid LoRaWAN Localization using Ensemble Learning

Abstract

Localization is an integral element in any Internet-of-Things (IoT) setup. Low Power Wide Area Network (LPWAN) technology such as LoRaWAN enables long-range communication with low-power consumption, and so, localization methods for outdoor applications have grown increasingly popular. Various ranging and fingerprinting techniques have been studied so far; however, fusing signal strength with time information remain to be an unexplored approach. In this study, an ensemble learning-based outdoor positioning algorithm utilizing hybrid data is designed with the goal of improving accuracy. An open access LoRaWAN dataset that offers both signal strength measurements and nanosecond precise timestamps is used and is split to train, evaluate, and test the algorithm that incorporates k Nearest Neighbors (kNN) method with the Random Forest Regressor (RFR). The kNN-RFR algorithm configured with the best parameters achieved a mean error of 332.63 meters and a median error of 193.63 meters.