Performance of vehicle generators classifying spatially patternized vehicles in traffic in developing countries

In heterogeneous traffic observed in developing countries, certain types of vehicles tend to travel together in small clusters, known as “Groups,” which can influence overall traffic flow and stability. Accurately reproducing such group behavior in traffic simulations is important for realistic modeling and for exploring ways to improve traffic flow. This study evaluated vehicle generators for microscopic traffic simulation that classify traffic entities into Groups and others (Remains) under heterogeneous and non-lane-based conditions. Three classifiers, i.e., vehicle generators, based on EDL and mEDL frameworks were compared using F1, F−1 (for Groups), and their product to assess the balance between detecting both categories. Classifier III achieved the most balanced performance despite being trained without explicit Group labels, indicating that it learned group-related patterns from indirect cues. While further improvements are needed for practical application, the proposed mEDLbased generator could contribute to more accurate heterogeneous-traffic simulations and provide a foundation for strategies that leverage Group behavior to enhance traffic flow.