Learning a spatial crash typology representation for analyzing and improving multimodal road safety in New England

The research project titled "Learning a spatial crash typology representation for analyzing and improving multimodal road safety in New England" aims to address the persistent issue of roadway crashes and fatalities by developing a comprehensive understanding of crash patterns and drivers. Leveraging data-driven spatial crash typology, this study seeks to analyze key patterns influencing roadway crashes, considering an array of factors including geography, topology, mode of transit, vehicle type, and driver behavior. The project will collect and integrate crash data from various New England states, applying machine learning and geospatial methods to develop a typology of census tracts based on crash characteristics. The research comprises six tasks, including data collection, dimensionality reduction, clustering, typology pattern analysis, model estimation for crash type prediction, and dissemination of findings. The analysis will utilize supervised learning methods to predict crash type classification of a given census tract, considering network topology, socioeconomic data, and travel behavior. The outcomes of the research will be shared through a published dataset, a dashboard for exploring typology, and model results, enhancing the usability and relevance for state agencies in crash monitoring and mitigation.