Analysing the effectiveness of unsignalized crossing infrastructure in improving pedestrian safety using multiple data-driven approaches

Abstract

This study investigates the effectiveness of unsignalized crossings to enhance pedestrian safety through a robust data-driven approach utilizing multiple machine learning models, including the statistical classifier Logistic Regression, Decision Tree, Random Forest, and Neural Network Multi-Layer Perceptron (MLP). While numerous studies have applied predictive models to traffic crash data, few have systematically analysed pedestrian crash severity at unsignalized crossings using multiple machine learning algorithms. By leveraging historical crash data from the UK's STATS19 database, key factors influencing pedestrian safety at unsignalized crossings were identified and analysed. The research highlights the superior predictive performance of Random Forest and MLP models, with accuracies of 84 % and 86 %, respectively, underscoring their capability to handle complex, nonlinear relationships in crash data. Feature importance analysis revealed critical determinants of crash severity. The findings emphasize the need for targeted interventions to mitigate crash severity of crash outcomes. Despite challenges like underreporting and data imputation biases, this study provides valuable insights into the role of infrastructure in pedestrian safety, offering a foundation for policy recommendations and future research on improving unsignalized crossing designs.