Bike-sharing systems for sustainable cities: Demand prediction using graph neural networks and spatial aggregation
Year: 2025
Type: article
Abstract: Demand forecasting in bike-sharing systems is crucial for efficient urban bike fleet management and enhanced user satisfaction. This study offers a comprehensive analysis of Warsaw’s bike-sharing system over a cycling season, incorporating temporal, spatial, environmental, and weather-related data. By aggregating station-level information into centroid-based cells defined on a kilometer grid, the approach effectively captures complex spatiotemporal interactions. Various Graph Neural Network (GNN) architectures were evaluated, with a broad range of hyperparameters tested to identify the most accurate predictive models. These models were assessed by their ability to forecast the volume of bike trips between stations, yielding results adaptable to complex urban contexts. Findings underscore the significance of weather, local infrastructure distribution, and temporal patterns in shaping bike usage. The resulting predictive models enable strategic decisions in station planning, bike redistribution, and overall fleet management, thereby increasing availability and improving system performance. Achieving R² scores above 0.91 and MAE below 0.37 highlights the high precision of these models. The introduction of an innovative spatial aggregation strategy, along with an extensive exploration of GNN architectures, marks a key contribution of this work. This methodology can be transferred to other major cities aiming to enhance their bike-sharing operations and foster sustainable urban mobility.
Source: Sustainable Cities and Society
Institution: Military University of Technology in Warsaw
Language: English
Subfield: Transportation
Field: Social Sciences
Domain: Social Sciences
Open Access status: hybrid
Estimated APC paid: $4,870
OpenAlex ID: W4410635362