Intersection crash analysis considering longitudinal and lateral risky driving behavior from connected vehicle data:...

Area of Science:

• Transportation Engineering
• Traffic Safety Analysis
• Machine Learning Applications

Background:

• Traditional intersection safety studies focus on macro-level data, neglecting micro-level driving behaviors.
• Connected Vehicle (CV) technology enables extraction of detailed driving dynamics.
• Lateral turning behaviors at intersections are critical for safety but have been understudied.

Purpose of the Study:

• To comprehensively analyze intersection driving dynamics by including both longitudinal and lateral behaviors.
• To address spatial heterogeneity and nonlinear effects in crash frequency prediction.
• To improve the accuracy of intersection crash prediction models.

Main Methods:

• Extraction of driving behavior features, including longitudinal movements and lateral turns, from CV data.
• Development of a novel spatial Machine Learning (ML) framework integrating nonlinear ML models (LightGBM) with geographically weighted regression.
• Training both global and localized ML models to capture average estimations and spatial heterogeneity.

Main Results:

• Inclusion of lateral turning behaviors significantly enhanced intersection crash frequency prediction accuracy.
• The proposed spatial ML framework integrating LightGBM outperformed traditional models (Random Forest, XGBoost, LightGBM, MLP) in RMSE, MAE, and R².
• Driving features exhibit nonlinear impacts and spatial heterogeneity; hard braking and acceleration influence rear-end crashes differently in urban vs. downtown areas, while hard left turns impact sideswipe and left-turn crashes in suburban areas.

Conclusions:

• Lateral turning behaviors are crucial predictors of intersection safety.
• Spatial ML frameworks effectively capture localized driving patterns and improve crash prediction.
• Understanding the context-specific (e.g., downtown, suburban) nonlinear impacts of driving behaviors is vital for targeted safety interventions.