From Coarse to Fine: Managing Temporal Granularity in Spatio-Temporal Data for Fine-Grained Traffic Prediction

Researchers propose STRP, a machine learning framework that predicts fine-grained traffic patterns from coarse-grained historical data, addressing a critical mismatch between how traffic data is stored and how it needs to be used. The solution combines tree convolution and inverse dilated convolution to efficiently model spatial and temporal dependencies, outperforming existing approaches while reducing computational overhead.

The article addresses a fundamental infrastructure challenge in spatio-temporal data systems: the tension between storage efficiency and prediction precision. Traffic management systems typically record observations at fixed intervals to minimize storage costs, yet downstream applications—from ride-sharing optimization to autonomous vehicle navigation—require predictions at much finer temporal resolutions. STRP bridges this gap through a novel dual-component architecture that enables accurate upsampling of coarse-grained data without requiring comprehensive fine-grained collection and storage.

This problem emerges from the practical constraints facing large-scale data systems. Maintaining fine-grained traffic observations across all locations and time periods would create exponential storage and preprocessing burdens. By developing methods to intelligently extrapolate from coarse data, STRP reduces infrastructure costs while maintaining prediction quality—a critical advancement for both public transportation systems and commercial logistics platforms.

The framework's practical impact extends across multiple industries. City planners and transportation authorities can optimize traffic signal timing with better predictions without upgrading sensor networks. Logistics companies gain improved route optimization capabilities. The interpretability offered through tree convolution makes model decisions auditable, important for infrastructure applications affecting public safety and resource allocation.

The benchmark performance across six datasets suggests broad applicability beyond traffic prediction. Similar granularity mismatch problems exist in weather forecasting, sensor networks, and financial data systems. Development of generalizable techniques for temporal upsampling could enable cost-effective data collection strategies across IoT and monitoring applications.

• →STRP enables accurate traffic predictions at fine temporal granularity using only coarse-grained historical data, reducing database storage and computation costs
• →Tree convolution and inverse dilated convolution components provide both efficiency and interpretability in spatio-temporal modeling
• →The framework supports flexible prediction modes (window-based and duration-based) to accommodate different data collection patterns
• →Experimental results demonstrate significant improvements in both accuracy and computational efficiency over existing state-of-the-art methods
• →The approach addresses a broader category of granularity mismatch problems relevant to IoT, sensor networks, and distributed data systems