This paper examines recent research in the field of public transportation, specifically focusing on the development of learning algorithms for predicting the behavior of trains and buses. However, it underscores the overlooked significance of having a clear and structured representation of data entities. To address this oversight, a relational model is proposed that captures the essential data fields specific to the subway system to enhance the learning process. The model undergoes validation through collaboration with a metro control center and domain experts. Furthermore, the study integrates this relational model into a hybrid approach that combines online and offline machine learning techniques. This approach effectively forecasts delays and passenger flow, thereby enabling informed decision-making and optimizing rail operations through a decision support system. The paper concludes by emphasizing the pivotal role of the proposed model in facilitating the selection of relevant variables for each learning problem.