Assessing the impacts of enriched information on crash prediction performance

Area of Science:

• Transportation Science
• Spatial Analysis
• Road Safety Engineering

Background:

• Effective road safety strategies rely on reliable data, which is often unavailable in developing countries.
• Limited data restricts predictive models to socio-economic and demographic variables, hindering accurate crash analysis.
• This data scarcity presents a dilemma, forcing difficult choices between inaction and implementing potentially ineffective measures.

Purpose of the Study:

• To demonstrate how enhanced explanatory variables and comprehensive datasets improve spatial crash prediction model performance.
• To model road traffic casualties as a function of richer data, including appropriate exposure variables.
• To compare model performance using available versus enriched datasets in São Paulo, Brazil, and Flanders, Belgium.

Main Methods:

• Development of spatial crash prediction models using Geographically Weighted Regression with Poisson distribution.
• Categorization of casualties into active (pedestrians, cyclists) and motorized transport modes.
• Comparative analysis of models using limited socio-economic/demographic data versus comprehensive datasets.

Main Results:

• Enriching datasets with supplementary information led to significant reductions in model error metrics (AICc and MSPE).
• Reductions in error were observed for both motorized transport (20-25%) and active transport (25-35%).
• The study highlights the impact of incorporating appropriate exposure variables and detailed transport mode data.

Conclusions:

• Comprehensive datasets and appropriate exposure variables substantially enhance the performance of spatial crash prediction models.
• Improved models can effectively identify accident hotspots and key contributing factors, guiding policy and resource allocation.
• Prioritizing data collection and utilizing enriched information can lead to more effective, cost-efficient road safety interventions.