Bridging land use and transport planning: An AI-enabled decision support system for new urban developments

Rapid urban expansion demands immediate land use decisions whose transport implications shape accessibility for decades. Yet traditional approaches lack transparent mechanisms to link controllable planning features (density regulations, network design standards, zoning policies) to observable mobility consequences, breaking the connection between what planners can control and what they need to predict. To address this fundamental challenge, we introduce Adaptive Dynamic Analysis for Predictive Transport (ADAPT), an explainable AI framework employing a two-stage architecture: Stage One learns categorical traffic patterns from 4397 established scenarios using conditional variational autoencoders, revealing that functional classification explains 38% of variance while density and spatial context contribute 28%. Stage Two transfers learned patterns to new developments through intelligent initialization and online adaptive learning. ADAPT advances land use-transport integration by linking planning features to observable mobility patterns through interpretable prediction. It provides three capabilities: feature importance estimates that show which planning characteristics predict traffic generation, context-dependent mechanisms that identify nonlinear thresholds across urban forms, and uncertainty-aware protocols that align infrastructure commitment with prediction confidence. Validated across 663 new development scenarios in three metropolitan areas (Los Angeles, San Diego, Tokyo), ADAPT achieves 17.9–35.1% performance improvements over state-of-the-art baselines while enabling evidence-based planning from initial approval stages.