Prediction of pedestrian crossing behaviour at unsignalized intersections using machine learning algorithms: analysis and comparison
Dungar Singh,Pritikana Das,Indrajit Ghosh
DOI: https://doi.org/10.1007/s12193-024-00433-0
2024-05-29
Journal on Multimodal User Interfaces
Abstract:The primary safety hazard at unsignalized intersections, particularly in urban areas, is pedestrian-vehicle collisions. Due to its complexity and inattention, pedestrian crossing behaviour has a significant impact on their safety. This study introduces a novel framework to enhance pedestrian safety at unsignalized intersections by developing a predictive model of pedestrian crossing behaviour using machine learning algorithms. While accounting for crossing behaviour as the dependent variable and other independent variables, the analysis prioritises accuracy and internal validity. Important feature scores for the different algorithms were assessed. The model results revealed that the arrival first of a pedestrian or vehicle, pedestrian delay, vehicle speed, pedestrian speed, age, gender, traffic hour, and vehicle category are highly influencing variables for analysing pedestrian behaviour while crossing at unsignalized intersections. This study found that the prediction of pedestrian behaviour based on random forest, extreme gradient boosting and binary logit model achieved 81.72%, 77.19% and 74.95%, respectively. Algorithms, including k-nearest neighbours, artificial neural networks, and support vector machines, have varying classification performance at every step. The findings of this study may be used to support infrastructure-to-vehicle interactions, enabling vehicles to successfully negotiate rolling pedestrian behaviour and improving pedestrian safety.
computer science, cybernetics, artificial intelligence