Enhanced Mobile Vehicle Tracking Based on LPWAN and AI for Logistics Applications

Accurate and cost-effective vehicle localization is essential for fleet management and logistics services. Although GPS-based tracking is widely used, it suffers from high power consumption and signal attenuation in transport vehicles, resulting in high cost and loss of precise location. This research proposes a machine learning-driven GPS-free localization approach using a Low-Power Wide-Area Network (LPWAN) communication technology i.e., LoRaWAN, which is already used to send location and other data like temperature and status of goods to the operator. The location is calculated using signal characteristics from a mobile vehicle transmitting data to multiple LoRaWAN gateways (GWs) and LoRaWAN Network Server (LNS). A LoRaWAN device records radio signal parameters like RSSI, SNR, spreading factor, and gateway parameters, with ground truth GPS coordinates for validation. The device's latitude and longitude serve as targets, while machine learning models predict location. Four experimental scenarios are evaluated: raw data, polynomial feature engineering, feature selection, and their combination. The performance is assessed via Mean Distance Error (MDE) and Cumulative Distribution Function (CDF) analysis. Results indicate that feature selection improves accuracy, while polynomial transformations further enhance the Gradient Boost, achieving 209.20 meters of localization accuracy. This work provides a practical and scalable solution for logistics applications, enabling low-power, cost-effective, and accurate vehicle tracking without GPS, where energy efficiency and robust localization are critical for operational efficiency.