Transportation mode estimation from large-scale human mobility data

As GPS-enabled devices become increasingly widespread, analyzing human mobility through recorded trajectories has become essential for developing intelligent transportation systems and promoting sustainable urban planning. However, these GPS datasets are often large-scale, high-frequency, and noisy, making it challenging to accurately classify transportation modes and detect transition points within trips that involve multiple transportation modes. To address this, we propose a deep learning–based framework that jointly estimates transportation modes and identifies mode transition points using a multi-output model built on a convolutional neural network (CNN) and bidirectional long short-term memory (BiLSTM) architecture. The framework incorporates an effective STAY/MOVE segmentation strategy to separate stationary and moving intervals, improving both mode recognition and transition detection. Rich kinematic features, such as speed, acceleration, direction changes, and turning behavior, are derived from raw GPS records (latitude, longitude, and timestamps) to capture both spatial and temporal mobility patterns. To handle the large dataset efficiently, a PySpark-based scalable inference pipeline is implemented for model inference, improving overall processing efficiency and reducing computation time. To evaluate generalizability, cross-province validation is conducted on a large-scale GPS dataset collected from GPS loggers and smartphones across multiple cities in China. The best-performing configuration achieves an F1-score of 96.8% and a recall of 96.8% for transportation mode classification, as well as an F1-score of 97.5% and a recall of 94.7% for transition point detection, successfully distinguishing walk, bike, car, bus, and train modes. These results demonstrate the robustness, scalability, and real-world applicability of the proposed approach across diverse urban contexts, highlighting its potential for large-scale mobility analytics and intelligent transportation applications.