Enhancing spatial-temporal demand prediction in transportation systems through region generation using soft clustering

Accurate spatial-temporal demand prediction is crucial for the effective service management of various transportation platforms, such as ride-hailing and micro-mobility services. While existing research primarily focuses on developing better prediction algorithms for demand, the relatively overlooked process of region generation is also important. In particular, predictions based on improperly defined regions can lead to poor predictive performance due to the modifiable areal unit problem. Therefore, it is essential to generate regions that reflect actual spatial-temporal demand patterns before training prediction models. This study proposes a region generation technique using a soft clustering approach, allowing spatial atomic units to belong to multiple clusters, unlike the conventional hard clustering method where each unit belongs to only one cluster. The proposed method selects spatial atomic units located at cluster boundaries and allows them to be part of adjacent clusters, thereby maintaining the geographic continuity of each cluster while overcoming the limitations of fixed boundaries that fail to capture the influence between neighboring clusters. Using three real datasets, the performance in demand prediction is evaluated based on regions generated by the proposed method and several comparison methods. The results show that the proposed method not only achieves the highest prediction accuracy but also exhibits the lowest variance in prediction accuracy across clusters.