Assessment of Route Choice Models for Dynamic Traffic Assignment using Microscopic Simulation

Travel forecasting models predict changes in the travel patterns and propose improvements. This study evaluates the parameters in route choice models, such as binomial, proportional, multinomial logit, and C-logit in microscopic simulation-based dynamic traffic assignment. The average Geoffrey E. Havers (GEH) index of each route choice model was tabulated when comparing the simulated flow of the junctions with the observed flow. The results indicated that the binomial model generally yields the lowest GEH value, but since the model does not consider the travel costs in the decision process, it is not suitable for traffic impact studies. As for the proportional and multinomial logit models, the K-Shortest Path (K-SP) value has greater impact on the assignment results. With a K-SP value of 1, the proportional and multinomial logit models generated the lowest GEH index when the alpha factor was set to 3.0 and the scale factor was set to 25, respectively. Lastly, for the C-logit model, the assignment results are more sensitive to the calibration of the scale factor and beta values compared to the K-SP and gamma factor. A lower GEH index is always observed for the scale factor of 25 and the combination with a beta value of either 0.1 or 0.15, regardless of the values of gamma and the initial K-SP. When comparing the calibrated models with the original model, the C-logit model showed higher deviations, whereas the logit and proportional models showed no significant differences. These findings highlight the importance of parameter calibration, apart from providing significant insights into route choice modeling, especially in replicating the real route choice behavior of motorists in Malaysia.