Do multivariate measures anticipate traffic congestion?

Abstract Traffic congestion generates substantial socioeconomic and environmental burdens, underscoring the need for reliable prediction to enable timely control interventions. This study investigates traffic jams from a critical transition perspective and assesses the predictive performance of 12 multivariate indicators across macroscopic continuum model-based and real-world data-driven scenarios. These multivariate indicators are typically based on critical slowing down (CSD) and are derived on a moving-window basis by integrating information across the three fundamental traffic variables: density, velocity and flow. The methodological framework comprises four components: sensitivity analysis, significance testing, robustness assessment and composite indicator (CI) construction. Theoretical analyses and data-driven tests show that the multivariate indicators are promising and hold substantial potential for congestion prediction. Careful selection of detrending strategies is important to remove the inherent non-stationarities in the data as their effect has been found to be non-trivial. Overall, the study bridges the application of nonlinear dynamics and the practical demands of traffic engineering. It employes universal multivariate measures that provide actionable tools for mitigating congestion and managing urban traffic efficiently.