Prediction of Signalized Intersection Delays Using Genetic Programming, XGBoost, and Clustering Approach

Abstract

Precise estimation of average control vehicle delay is a critical parameter for the efficient operation and optimization of signalized intersections. This paper presents predictive models for estimating delays at signalized intersections that are developed by using Genetic Programming (GP) and Extreme Gradient Boosting (XGBoost). The combination of GP's ability to derive interpretable mathematical expressions and XGBoost's reliability in handling nonlinear relationships provides a comprehensive framework for accurate delay prediction. In order to capture data-driven patterns, the dataset was separated using k-means clustering and grouped-based segmentation. Both six k-means clusters and four different groups were compared with traditional delay models (HCM, and Ak & ccedil;elik). The results indicate that the models demonstrated good performance to estimate delays at signalized intersections under varying saturation degrees, reflecting different traffic volumes. These findings suggest that machine learning-based delay estimation models can significantly advance both theoretical research and practical applications in traffic management for signalized intersections.