Files
Abstract
Traffic safety analysis has traditionally been conducted at separate spatial levels, macro (areawide) and micro (location-specific, such as intersections or road segments), each with distinct methodologies and limitations. This study develops a multi-scale predictive framework that integrates these levels to improve crash frequency predictions and better prioritize intervention areas. Focusing on Georgia’s counties, the framework captures spatial dependencies and interactions in crash occurrences by combining statistical and machine learning (ML) techniques. At the macro level, Negative Binomial regression models county-wide crash frequencies based on demographic, land use, policy, and traffic exposure factors. At the micro level, advanced ML models analyze segment- and intersection-level crash data, incorporating roadway geometry, traffic flow, and environmental characteristics. To bridge these levels, the framework employs hierarchical modeling, spatial aggregation/disaggregation techniques, and cross-level feature integration. Results show that this integrated approach improves prediction accuracy and enhances the identification of critical safety factors compared to standalone models. The findings provide actionable insights for policymakers and transportation planners, supporting data-driven safety interventions to reduce crash frequency and severity across transportation networks.