Graph Neural Networks for Predicting Urban Service Demand from Sparse Mobility Signals

Abstract

Predicting the demand for services in urban cities has become one of the main challenges in large-scale planning for smart cities, as traditional approaches rely on population movement data in terms of direction and continuity. However, these data are often incomplete or limited due to technical and cost constraints or in terms of specificity. In this study, an intelligent system was presented to predict service demand in urban cities, relying on neural networks in the presence of limited population movement signals. In this system, the city was represented as a graph where nodes represent urban areas and edges represent spatial relationships between them. The proposed model allows benefiting from the spatial interconnections between areas to compensate for the direct lack of data. The results obtained show that the model based on neural networks achieves ordinary accuracy and better stability, especially compared to the traditional model in areas suffering from data scarcity, where the regression accuracy was R2 = 0.8.