A Just-In-Time Framework for Routing-Oriented Traffic Prediction

Traffic prediction plays a crucial role in urban transportation systems, yet existing methods face challenges in achieving real-time performance when handling large-scale road networks. This paper introduces a novel Just-In-Time Traffic Prediction framework that integrates traffic condition with routing queries for efficient localized predictions in multi-query urban environments. Unlike traditional approaches that perform global predictions across entire networks, our framework partitions the road network into non-overlapping small regions and selectively updates traffic conditions based on query demands. Specifically, we propose three key components: (i) a Search Space Estimation (SSE) model that reformulates search space determination of routing queries as a binary classification task to accurately identify the searched regions; (ii) a Region-based Traffic Speed Prediction (RTSP) model that incorporates the temporal validity of speed profiles in adjacent regions and comprehensive spatio-temporal features for precise region-based traffic prediction; (iii) a Global Region Prediction Scheduling that efficiently coordinates the SSE and RTSP models to maintain up-to-date traffic data for running queries while minimizing computational overhead from both spatio and temporal dimensions. Experimental results on real-world road networks demonstrate significant improvements in both effectiveness and efficiency compared to state-of-the-arts.