(2024-03-05) Maia, Pedro Lopes; Medeiros, Victor Wanderley Costa de; http://lattes.cnpq.br/7159595141911505; http://lattes.cnpq.br/2161981667043569
With the proliferation of the use of IoT technologies, efficient solutions in terms of battery usage and applicability for device positioning have become increasingly necessary due to the demand for location-based services. In this context, signal-based localization techniques, such as fingerprinting, represent a very appropriate solution as they meet the requirements of these applications. In this study, a public dataset containing RSSI values from a LoRaWAN network was used to create machine learning models to evaluate their effectiveness in positioning LoRa devices, offering an alternative to GPS, which due to the high power consumption of device batteries, in many cases, is not viable for IoT systems. After evaluating hyperparameters and applying appropriate methodologies for each algorithm studied, a model was obtained capable of making predictions with an average error of 301.34 meters and a median of 164.26 meters.
