AdaptiMOD: A Hybrid Deep Learning and Multi-Agent Reinforcement Learning Framework for Sustainable Multimodal Transportation in Developing Economies

Developing economies face unique urban mobility challenges: 3.5 billion people rely on heterogeneous informal transit systems lacking real-time optimization, while infrastructure constraints prevent the deployment of conventional AI solutions. I present AdaptiMOD (Adaptive Multimodal Optimization with Deep Reinforcement Learning), a novel hybrid framework combining spatio-temporal Graph Neural Networks (ST-GNN) for demand prediction, federated Multi-Agent Reinforcement Learning (MARL) for distributed transit coordination, and diffusion-based generative models for scenario synthesis in data-scarce environments. My framework addresses three critical gaps: (1) transfer learning from data-rich to data-scarce cities achieving MAE of 3.18 with only 30% training data (comparable to 2.93 with full data within 8.5% performance gap), (2) edge-optimized models requiring