Tabu-Search-Based Combinatorial Subset Selection Approach to Support Investigation of Built Environment and Traffic Safety Relationship

Abstract

Traffic crashes are a leading cause of death globally, with an increasing rate in urban areas. Thus, this study focuses on the relationship between built environment (BE) and traffic safety (TS), by constructing a relationship model using BE variables. The aim of this paper is to determine the best subset of BE variables through a generalizable methodology. The BE is operationalized through the D-classification (e.g., density, diversity, and design), and various datasets are collected from different agencies. TS is operationalized through motor vehicle involved (MOT) and vulnerable road user (VRU)-involved crash frequencies at the zonal level. A preliminary GIS-based process is conducted to associate the crash data at the census block group (BG) level, followed by examining the BE-TS relationships through a series of negative binomial models optimized for subset selection. The model generation is performed automatically by an embedded Tabu Search procedure. Two case studies are presented: a single-county case (Leon County, Florida, U.S.) and a tri-county case (Miami-Dade, Broward, and Palm Beach Counties, Florida, U.S.). Results show that some BE variables such as total population, age of housing stock, number of bus stops, and traffic volume have consistently positive relationships with crash occurrences. In contrast, several factors show varying effects by crash type or location. For example, motorized mode percentage has a negative relation with crash occurrences in the single-county case whereas it is insignificant for the tri-county case where the non-motorized mode percentage has a positive effect on crash occurrences.