Future travel demands should be predicted accurately in order to plan, make operational decisions, and manage urban public transportation systems. The success of the developed prediction model will directly affect the success of the transportation plan. Many factors, such as day of the week, weather, whether there is a large organization in the city, whether schools are open, affect the demand for urban public transportation.
Organizations such as celebrations, festivals, sports competitions, or changes in the weather may cause a different travel demand than expected. The unexpected increase in travel demand makes it difficult to manage the transportation system. Daily travel demand predictions should be considered when making many operational decisions, such as arranging the frequency of services and determining the number of personnel to serve. Trip data of public transportation systems can be obtained easily by utilizing smart transportation cards. This large-scale dataset allows modelling the relationship between the above-mentioned factors and public transport usage demand using deep learning techniques.
This paper presents a comprehensive study on the development of a daily passenger demand prediction model for rail systems using deep learning techniques. The study incorporates a wide range of external factors, including weather conditions, day of the week, public holidays, and the occurrence of specific events such as football matches, to create a prediction model. Various deep learning models with different variable sets were developed using the daily travel data of the 2019 Istanbul M2 Yenikapı – Hacıosman metro line. The impact of various external factors on travel demands were systematically examined by assessing the prediction performances of five different deep learning models created with different set of variables.

: public transit, demand prediction, deep learning, railway systems