Pedestrian safety at signalized intersections: Spatial and machine learning approaches

Abstract

IntroductionThe major goal of the present research is to determine hotspot areas by the generation of a geospatial model and develop a model associated with pedestrian-vehicle crash injuries (severe, moderate, slight) at signalized intersections in Erzurum, Turkey. MethodologyThis study used the comprehensive algorithm in Artificial Neural Network (ANN). Data from 197 crashes injury (2015–2019) at 57 intersections depending on the mix of variables such as driver, road and vehicle characteristics, and environment data were collected. ResultsWithin the four candidate models, the first one including pedestrian density, level of education, traffic congestion, type of vehicle, presence of bus stop, age, and gender had the lowest RMSE and MAE values and the greatest R2 value. Lastly, sensitivity analyses were conducted to evaluate the impact of independent parameters. ConclusionsThe importance of the study lies in the expected outcomes to assist the experts to address the pedestrian-vehicle crash risk factors by conducting appropriate countermeasures for facilities management/improvement.