Project Summary
California cyclist KSI crashes increased 35% and pedestrian KSI increased 20% between 2020 and 2024. This project addresses that crisis by developing a statewide machine learning pipeline to identify predictors of pedestrian and cyclist crash risk across California. Using SWITRS/TIMS crash records (2022–present), this research extends the PI’s prior random forest methodology to a statewide scale, linking crashes to OpenStreetMap infrastructure attributes — road class, speed limit, bicycle facility type, intersection configuration, and Level of Traffic Stress — alongside exposure estimates from direct demand models (DDMs) trained on statewide count data, ACS demographics, NOAA weather, Smart Location, and PeopleForBikes network metrics. DDMs produce segment-level bicycle and intersection-level pedestrian exposure estimates, enabling crash rate calculation rather than raw counts. Models generate spatially explicit crash risk and injury severity predictions for each mode, with variable importance outputs identifying which infrastructure conditions most strongly predict elevated risk. Outputs are translated into engineering opportunity maps prioritizing locations where targeted interventions have the highest safety benefit. The pipeline is fully scripted for annual re-execution as new SWITRS data are released, giving agencies a continuously updated decision-support tool.
