Prototype Models for Predicting Vehicle Types Generated in Heterogeneous Traffic Simulation

Traffic in emerging countries often comprises various types of vehicles, i.e., it is heterogeneous. Previous studies have proven that the order of the vehicle types affects the properties of traffic, and the present authors confirmed that grouping behaviors exist in the order of field traffic. It is possible to incorrectly evaluate traffic if one randomly generates vehicles in a traffic simulation. In this study, to accurately replicate such spatial patterns, we compared the performances of prototype models for prediction the types of vehicles generated in a simulation area. Prototype models, including Gaussian-process (GP) generators and evidential deep learning (EDL) generators, increase prediction performance. However, different behaviors for random and patterned sections in traffic have also been observed. This study contributes not only to developing accurate traffic simulation models, but also to understanding the internal emerging structures of heterogeneous traffic.