Short-Term Traffic Speed Prediction Integrating Traffic Flow Theory for Enhanced Urban Congestion Mitigation

Abstract Effective mitigation of urban traffic congestion requires accurate speed prediction. It also requires a proactive understanding of how traffic states evolve. Existing deep learning models often work as black boxes. They do not capture the underlying dynamics of traffic flow transitions, which limits their use for practical congestion management. To address this gap, this study proposes a theory guided deep learning framework that integrates traffic flow theory with short-term speed prediction. The K-means clustering algorithm was used to classify traffic into three states based on the fundamental speed and flow diagram. These states are free-flow, transition, and congested. The state label was then added as an input feature to nine deep learning models. The method was validated on thirty-eight congested freeway segments in South Korea. On average, the improvement in prediction accuracy was modest. However, the benefit was concentrated in worsening regime transitions, in which the state feature reduced errors approximately four times more than in steady conditions. A statistical test confirmed that this effect was significant in most cases. The framework is therefore most useful when the flow begins to break down, which is when an early warning matters most. This allows operators to act early with measures such as variable speed limits before congestion fully develops, shifting traffic management from reactive monitoring toward proactive control.