Detecting Dynamic Communities in Vehicle Movements Using Ant Colony Optimization

Abstract

Detecting dynamic community structure in vehicle movements is helpful for revealing urban structures and human mobility patterns. Despite the fruitful research outcomes of community detection, the discovery of irregular-shaped and statistically significant dynamic communities in vehicle movements is still challenging. To overcome this challenge, we developed an evolutionary ant colony optimization (EACO) method for detecting dynamic communities in vehicle movements. Firstly, a weighted, spatially embedded graph was constructed at each time snapshot. Then, an ant-colony-optimization-based spatial scan statistic was upgraded to identify statistically significant communities at each snapshot by considering the effects of the communities discovered at the previous snapshot. Finally, different rules defined based on the Jaccard coefficient were used to identify the evolution of the communities. Experimental results on both simulated and real-world vehicle movement datasets showed that EACO performs better than three representative dynamic community detection methods: FacetNet (a framework for analyzing communities and evolutions in dynamic networks), DYNMOGA (dynamic multi-objective genetic algorithm), and RWLA (random-walk-based Leiden algorithm). The dynamic communities identified by EACO may be useful for understanding the dynamic organization of urban structures.