Abstract

Using artificial intelligence (AI) techniques, this thesis illustrates how public transportation can be optimized in line with smart city visions. Deep learning models can be used to analyze complex datasets in order to predict demand, streamline operations, and adapt dynamically to real-time conditions. The research focuses on two key components: first, a Graph Neural Network (GNN) with probabilistic node embeddings is used to predict passenger flow in bus networks, improving accuracy and optimizing resource allocation. This method is validated using Automated Passenger Counting (APC) data from Laval, Quebec. The second component employs a Reinforcement Learning (RL) algorithm to enhance ride-matching and vehicle routing in on-demand shared mobility services, particularly addressing the first-mile problem in suburban areas. Sensitivity analyses confirm the adaptability and effectiveness of both approaches, showcasing the potential of AI to improve transportation efficiency in diverse urban and suburban contexts.