데이터마이닝과 베이지안 기법을 활용한 소규모 지역의 부주의 운전에 의한 교통사고 위험 모형 구축

Abstract

This study aims to identify influential factors affecting the relative risks of vehicle crashes related to distracted driving at small area-levels (i.e., census block group) in Northern Ohio, U.S. To this end, we utilize data mining techniques to analyze publicly available data from multiple sources, which include spatial, sociodemographic, and environmental features alongside crash data. We then develop several generalized linear mixed models using a full Bayesian hierarchical formulation. The selected optimal model indicates that factors such as gross activity density, landuse mix, and density of intersection appear to increase the relative risks of vehicle crashes due to distracted driving, while population density contributes to reducing these risks. In addition, we observe substantial random fluctuation of residuals originating from both block-group and census tract level variability across the study region. These findings allow us to help build regional transportation safety policies to mitigate negative consequences of distracted driving.