Kidando, Emmanuel (Washkewicz College of Engineering - Cleveland State University), Balyagati, Philip (Cleveland State University), Ngereza, Abdul (Cleveland State University), Kutela, Boniphace (Texas A&M Transportation Institute), Kalambay, Panick (University of Washington Tacoma), Kitali, Angela E. (University of Washington Tacoma)

Integrated Multi Regime and Gaussian Processes Model for Calibrating Traffic Fundamental Diagram

Scheduled for presentation during the Regular Session "Traffic Theory for ITS" (FrBT12), Friday, September 27, 2024, 14:30−14:50, Salon 20

2024 IEEE 27th International Conference on Intelligent Transportation Systems (ITSC), September 24- 27, 2024, Edmonton, Canada

This information is tentative and subject to change. Compiled on December 26, 2024

Abstract

Single-regime and multi-regime regressions are early models used to calibrate the relationships among traffic variables, i.e., speed, density, and flow. Recent advancements in computational power have enabled the development of non-parametric models based on machine learning, significantly enhancing the estimation accuracy of these relationships. However, using non-parametric assumptions often limits the interpretability of the fitted models. This study aims to enhance a non-parametric model, particularly the Gaussian Process Regression (GPR), by integrating it with the two-regime (TR) model to reduce estimation bias and improve model interpretability. The integrated TR and GPR (ITR + GPR) model reduced estimation bias, especially in extreme regions of the occupancy-speed relationship, such as in areas with low and high speeds. Furthermore, the study showcased the application of the proposed framework in clustering congested and free-flow regimes using the calibrated membership probabilities. The results highlight the potential of the integrated model in accurately capturing complex traffic data characteristics while providing improved interpretability.