Spatiotemporal analysis of urban traffic crash risk using a bagging-optimized dynamic mode decomposition framework

Area of Science:

• Urban planning and transportation safety
• Data science and machine learning applications
• Spatiotemporal analysis of risk dynamics

Background:

• Understanding urban traffic crash dynamics is crucial for public safety.
• The COVID-19 pandemic significantly altered urban mobility and traffic patterns.
• Accurate short-term crash prediction remains a challenge in complex urban environments.

Purpose of the Study:

• To analyze the spatiotemporal evolution of traffic crash risk in Manhattan.
• To enhance one- to seven-day-ahead traffic crash prediction using a novel BOPDMD framework.
• To investigate differences in crash patterns between pre-pandemic and pandemic years.

Main Methods:

• Utilized two years of daily Manhattan crash data (2019-2020) aggregated at the neighborhood level.
• Applied a Bagging Optimized Dynamic Mode Decomposition (BOPDMD) framework for prediction and spatiotemporal analysis.
• Compared BOPDMD performance against standard Dynamic Mode Decomposition (DMD) and baseline models for 1-7 day ahead forecasts.

Main Results:

• BOPDMD demonstrated superior accuracy (lower MAE and RMSE) in crash prediction compared to baseline models.
• Spatiotemporal mode analysis revealed distinct crash dynamics between 2019 (stable, periodic) and 2020 (disrupted, transient).
• The framework successfully captured pandemic-induced shifts in crash risk allocation and temporal patterns.

Conclusions:

• BOPDMD offers accurate short-term crash forecasting and interpretable insights into urban risk dynamics.
• The study highlights the impact of external disruptions (like pandemics) on traffic safety.
• Findings support proactive urban safety management through targeted interventions for persistent and emergent crash risks.