This paper proposes a LoRa-based positioning and tracking method designed for large-scale outdoor environments. We develop a positioning algorithm that integrates Kalman filtering with a neural network. The algorithm first applies Kalman filtering to preprocess the RSSI and SNR values of LoRa signals, effectively reducing noise and smoothing the data to enhance signal reliability. For position prediction, we construct a multi-input LSTM neural network model that fuses static features (such as RSSI and SNR) with temporal features (such as historical location and time differences) to accurately estimate the device’s location. We evaluate the proposed algorithm on a public dataset collected by three LoRaWAN gateways, conducting a comprehensive performance analysis in terms of positioning accuracy and error distribution. Experimental results demonstrate that the algorithm reduces the average positioning error to 50 m, significantly outperforming existing state-of-the-art neural network-based positioning methods.