Demand and supply are crucial elements of the ride-hailing business. After the evolution of the GPS supported mobile-based ride-hailing systems, hotspots detection in a spatial region is one of the most discussed topics among the urban planners and researchers. Due to the high non-linearity and complexity of traffic flow, traditional methods cannot satisfy the requirements of prediction tasks and often they neglect the dynamic constraint of a spatial region. To address this issue, this research considered the road network graph and transform hotspots detection problem into a node-wise decision-making problem and extracted subgraphs as hotspots. In this paper, the authors propose a graph neural networks enable reinforcement learning agents to learn the dynamic behaviour of a road network graph and use it for a subgraph extraction in a road network graph. The Graph Neural Networks (GNNs) can extract node features like the pickup requests and events in the city and generate the subgraphs by stacking multiple neural network layers. Experiments show that the proposed model effectively captures comprehensive spatio-temporal correlations and outperforms state-of-the-art approaches on real-world taxi datasets. © 2022, The Author(s), under exclusive license to Springer Nature Switzerland AG.