Research: Spatiotemporal Deep Learning for Urban Mobility Pattern Discovery Using Large...

Research question: How can spatiotemporal deep learning models accurately capture and predict urban mobility patterns from large-scale GPS traces, and what are their limitations compared to classical approaches?

Urban mobility modelling is crucial for city planning, transportation infrastructure, and understanding human movement dynamics. The proliferation of GPS-enabled devices has resulted in massive datasets capturing fine-grained mobility patterns across major cities worldwide.

Despite extensive work on urban mobility prediction, existing models often fail to fully exploit the spatial and temporal dependencies inherent in GPS trace data. Classical statistical approaches and traditional machine learning methods may overlook complex interactions and non-linearities present in real-world movement.

This project proposes to investigate advanced spatiotemporal deep learning architectures—such as Graph Neural Networks and Temporal Convolutional Networks—on large-scale, publicly available GPS datasets. The methodology includes rigorous benchmarking against classical methods, evaluating predictive accuracy, interpretability, and scalability.

The expected contribution is a systematic comparison of model effectiveness and limitations, providing actionable insights for urban analytics. The findings will inform future research in mobility prediction, smart city applications, and reproducible analytics in urban data science.