Defining accuracy benchmarks for freeway traffic simulations in support of highway operations and planning

Accurate calibration of traffic simulation models is essential for replicating observed traffic conditions, and subsequent optimization of decision-making processes and targeted investments in transportation infrastructure. This study applies a genetic algorithm (GA) to optimize key parameters of the car-following model for a basic freeway segment in California, aiming to minimize the error between simulated and observed traffic data. Outputs generated during GA iterations were analyzed using paired T-tests and Wilcoxon signed-rank tests to compare simulated speed and flow against ground truth data. Accuracy for each sample was matched to its corresponding P-value, revealing a clear trend: when accuracy levels exceeded 80%, P-values for both speed and flow consistently rose above 0.05. This indicates that the simulated outputs became statistically indistinguishable from the observed field data after 80% accuracy. These findings demonstrate that combining statistical significance with accuracy metrics can effectively guide calibration processes and establish thresholds for acceptable simulation accuracy, contributing to a robust framework for traffic simulation studies.