Data-Driven Bike-Share Ridership Prediction and Network Optimization

Abstract

Shared micro-mobility systems, particularly station-based bike-sharing networks, have become key components of urban transportation, yet their planning remain challenged by spatial and technological complexities. This dissertation develops integrated models for ridership prediction, station placement optimization, and electrification planning to address these challenges.

First, a customized Graph Neural Network framework using GraphSAGE is introduced for station-to-station ridership prediction, integrating network topology, sociodemographic features, and station attributes. Applied to Toronto, the model outperforms linear, spatial, and tree-based benchmarks, demonstrating its ability to capture latent dependencies and support demand-responsive planning. Second, a continuum approximation model is proposed for station placement optimization, using a force-based algorithm that balances attraction from demand centers with inter-station forces. This ridership-driven approach departs from conventional accessibility methods by directly aligning locations with demand. Applied to Vancouver, the model reveals optimal spacing patterns and highlights strategies for ridership-driven network expansion under varying demand conditions. Third, the dissertation extends infrastructure planning to electrified systems by introducing a two-dimensional Markovian state-of-charge framework for e-bikes. A heuristic charger deployment algorithm, enhanced by a single-pooling state approximation, maximizes expected ridership and identifies high-impact charging locations, achieving near-optimal performance in case studies from Pittsburgh, Vancouver, and San Francisco. Finally, the models are integrated into a web-based, GIS-enabled decision-support tool that combines predictive, prescriptive, and descriptive analytics to enable scenario-based planning. Demonstrations in Toronto and Vancouver illustrate the tool’s scalability and practical value. This research advances methodological foundations and practical tools for developing resilient, data-driven, and electrified bike-sharing networks across diverse urban environments.