Predicting Pedestrian Crossing Behavior at Urban Intersections Using Machine Learning Techniques : Evidence from Hyderabad, India

Abstract

Rapid urbanization and increasing traffic volumes have intensified conflicts between vehicles and pedestrians at urban crossings, making pedestrian safety a critical concern. This study investigates pedestrian crossing behavior using machine learning techniques at five major intersections in Hyderabad, India: Uppal, Dilsukhnagar, LB Nagar, Chaitanyapuri, and Meerpet. Data were collected from 400 pedestrians through a structured questionnaire covering demographic characteristics, trip attributes, traffic conditions, signal compliance, and perceptions of pedestrian facilities. Logistic Regression, Support Vector Machine (SVM), Random Forest, Gradient Boosting, and Artificial Neural Network (ANN) models were developed and evaluated using accuracy, precision, recall, and F1-score. The results demonstrate that machine learning models effectively predict pedestrian crossing behavior under diverse urban traffic conditions. Among the evaluated models, Random Forest achieved the highest prediction accuracy (82.19%), followed by Gradient Boosting (79.45%) and SVM (76.71%). Age, gender, traffic volume, signal waiting time, trip purpose, and the quality of pedestrian infrastructure were identified as the most influential factors affecting crossing decisions. Safety, accessibility, and walking distance also significantly influenced the choice of crossing facilities. The findings provide valuable insights into pedestrian decision-making and support data-driven strategies for enhancing pedestrian safety, optimizing signal operations, and improving sustainable urban mobility planning.