A Case for Data-driven Urbanism: Graph Machine Learning-Based Approaches for Transit-Oriented Development Decision Support

Rapid advances in open‐source urban data and machine‑learning have outpaced the decision‑support tools available to planners tasked with delivering transit‑oriented development (TOD). This thesis confronts that gap by asking: How can a data‑driven design decision‑support framework be implemented for TOD using graph‑powered machine learning? Adopting an Action Design Research (ADR) stance, the study wove six build–intervene–evaluate cycles into everyday practice, coupling iterative tool‑making with systematic academic review. These cycles culminated in two pivotal graph‑based artefacts. First, a Neo4j database unifies New York City cadastral, land‑use, taxation and POI layers; label‑propagation traverses the land‑use–transport network to rank parcels by redevelopment feasibility and acquisition practicality. Second, a graph‑neural network (GNN) supplies real‑time TOD guidance based on the typology study of 311 New South Wales stations, returning instant Node-Place-Design performance scores to designers. The implementations demonstrate how graph analytics can translate heterogeneous urban data into actionable, practice‑ready insight. The thesis contributes (i) a pragmatist definition of data‑driven design for urbanism, (ii) an extensible graph‑based data model linking land, transport and design variables, and (iii) evidence that GNNs can move beyond traffic forecasting to scenario‑sensitive urban design analytics. These findings equip practitioners and researchers with a scalable workflow for embedding AI into TOD planning, advancing the broader agenda of computationally‑supported, sustainable city‑making.