AI-Enabled Land-Use–Transport Decision Support for Bike-Sharing Expansion: Reproducibility Package

Abstract

Code, Colab notebooks, configurations, trained checkpoints and derived results for the article "AI-Enabled Land-Use–Transport Decision Support for Bike-Sharing Expansion: Spatial Transfer, Station-Mediated Accessibility, and Policy Frontiers" (under review in Transportation Research Part A: Policy and Practice, special issue on AI for land-use and transport). The study predicts sparse station-to-station bike-sharing flows in Warsaw with a hurdle graph neural network and hurdle MLP, hurdle GLM and gravity PPML benchmarks; tests transfer to unseen stations with a strict spatial holdout; inserts candidate stations into a frozen network; measures distinct-station opportunity-union accessibility weighted by GHS-POP population; and selects station portfolios on exact CP-SAT demand–accessibility frontiers with expansion, interaction and sensitivity analyses. Contents: stage A – the frozen v4.5 experiment (source modules, notebooks, checkpoints, complete results); stage B – GHS-POP population post-processing; stages C–F – revision analyses (revision experiments, strict spatial holdout and expansion-constrained frontiers, distinct-station correction of the accessibility baseline, analyses requested in review). Each stage includes its notebooks, code and complete result archives. Verification: scripts/validate_package.py (Python standard library only) checks the SHA-256 of all archives and file manifests and recomputes 67 values reported in the article from the archived tables. docs/MANUSCRIPT_MAPPING.md links every table and figure to its source files. Input data: Warsaw bike-sharing datasets on Mendeley Data (https://doi.org/10.17632/kzvdgfzk4w.1 and https://doi.org/10.17632/pxp72h4wdg.1, CC BY 4.0), GHS-POP R2023A (European Commission, Joint Research Centre) and OpenStreetMap (ODbL). Licences: code MIT; derived results, tables and figures CC BY 4.0; OSM-derived network cache ODbL 1.0.