Graph-Based Deep Learning and Multi-Source Data to Provide Safety-Actionable Insights for Rural Traffic Management

In rural arterial networks, poor sensor coverage, high vehicle speeds, and intricate traffic dynamics make Traffic State Estimation (TSE) an essential task. The intricacies of rural surroundings are not adequately captured by traditional TSE approaches, which rely on single-source data like loop detectors and GPS. This results in safety hazards like over speeding, queue spillback, and short headways. This study presents a novel strategy to overcome these issues by fusing sophisticated deep learning models with data from several sources. By combining a Graph Attention Temporal Convolutional Network (GAT-TCN) with traditional Kalman Filter (KF) variations (Extended, Unscented, and Sliding Window), we suggest a hybrid architecture. With its ability to capture both multi-resolution temporal dynamics and dynamic spatial dependencies, the GAT-TCN model performs noticeably better than conventional techniques in terms of Mean Absolute Error (MAE) and Root Mean Square Error (RMSE). By combining loop detector data and Bluetooth trip durations, empirical validation on a real-world rural toll route shows that the GAT-TCN improves safety by enabling early detection of important occurrences like over speeding and queue spillback and produces more accurate traffic projections. The findings demonstrate how combining multi-source data with state-of-the-art machine learning algorithms can enhance rural areas’ transportation efficiency and safety. The findings demonstrate how combining multi-source data with state-of-the-art machine learning algorithms can enhance rural areas’ transportation efficiency and safety. This study offers a scalable framework for proactive rural traffic management, marking a departure from conventional traffic status estimation in favor of safety-actionable insights.