Abstract
This study presents a theoretical foundation for a hybrid machine-learning model designed for real-time adaptive passenger flow prediction in public transportation systems. By integrating both historical and real-time data, the model aims to enhance demand forecasting accuracy and support optimized resource allocation. The proposed model combines Long Short-Term Memory (LSTM) networks for time-series dependencies with Graph Convolutional Networks (GCNs) to capture spatial relationships, enabling dynamic adaptation to changing patterns in passenger flow. Although this study is theoretical, anticipated practical applications include improved scheduling efficiency and responsiveness in transit systems, contributing to sustainable resource management. Key challenges, such as computational complexity and scalability for large datasets, are discussed alongside potential gains in prediction accuracy over traditional models, positioning this approach as a promising step forward in adaptive transit systems.