Machine Learning Approach in Calibrating VISSIM Microsimulation Model for Mixed Traffic Conditions

Authors

  • Ifratul Hoque Department of Civil Engineering, Bangladesh University of Engineering & Technology, Bangladesh
  • Tanmoy Nath Ananda Department of Civil Engineering, Bangladesh University of Engineering & Technology, Bangladesh
  • Parvez Anowar Department of Civil Engineering, Bangladesh University of Engineering & Technology, Bangladesh
  • Adeeba Naz Department of Civil Engineering, Bangladesh University of Engineering & Technology, Bangladesh
  • M Neaz Murshed Department of Civil Engineering, Bangladesh University of Engineering & Technology, Bangladesh

DOI:

https://doi.org/10.3329/jes.v16i1.82663

Keywords:

PTV VISSIM, Latin Hypercube Sampling, Cluster analysis, Calibration, Wiedemann 99

Abstract

Traffic Simulation has empowered transportation engineers by providing a means of visual interpretation for real-life traffic conditions. PTV VISSIM is a well-known microsimulation software used to analyze and predict traffic operations and behavior by considering factors such as lane configuration, traffic composition, transit stops, etc. A non-laned-based heterogeneous traffic stream characterizes the urban traffic system of Dhaka. This makes it burdensome to calibrate and validate VISSIM models to reflect field-obtained traffic flow. To calibrate VISSIM-developed simulation models, Weidemann 74 and 99 car-following models are widely adopted. These car-following models and other movement parameters, such as lateral movement and lane-changing behavior parameters, are usually adjusted to calibrate the microsimulation model. This study aims to develop a new approach using sampling and machine learning to calibrate the Weidemann 99 car following model parameters in VISSIM microsimulation software for mixed traffic conditions. A portion of Abdul Gani Road, which represents the typical characteristics of the traffic system of Dhaka, was chosen to be the epicenter of this study. Latin Hypercube Sampling has been used to generate the number of combinations required to properly explore the effects of the ten calibration parameters of the Weidemann 99 car following model on the validation accuracy. The validation accuracy has been measured by using the GEH statistic. A total number of 500 simulations were generated, and from these 500 simulations, 37 combinations were obtained to have acceptable GEH values, which is generally considered to be less than 5%. These combinations were further analyzed using a k-means clustering algorithm to generate the centroid line of the acceptable parameter combinations. A sensitivity analysis was conducted using the obtained simulation dataset to determine the impact of changing values of the parameters on traffic flow. The findings of this study will aid future traffic simulation researchers by providing them with a guiding framework in calibrating VISSIM simulation models for mixed traffic conditions similar to Dhaka.

Journal of Engineering Science 16(1), 2025, 21-30

Downloads

Published

2025-07-02

How to Cite

Hoque, I., Ananda, T. N., Anowar, P., Naz, A., & Murshed, M. N. (2025). Machine Learning Approach in Calibrating VISSIM Microsimulation Model for Mixed Traffic Conditions. Journal of Engineering Science, 16(1), 21–30. https://doi.org/10.3329/jes.v16i1.82663

Issue

Section

Articles