Light rail passenger demand forecasting by artificial neural networks

The success of strategic and detailed planning of public transportationhighly depends on accurate demand information data. Short-term forecasting isthe key to the success of transportation operations planning such astime-tabling and seat allocation. This study adopts neural networks to developshort-term passenger demand forecasting models to be used in operationalmanagement of light rail services. A multi-layer perceptron (MLP) model ispreferred due to not only its simple architecture but also proven success ofsolving approximation problems. For eliminating the significant seasonality intime slots, each time slot is handled independent of the others, and anartificial neural network based on daily data is developed for each. Regardingto the 74 different time slots, 74 different neural networks are trained byhistory data. Three illustrative examples are demonstrated on one of the timeslots and performance of the forecast models are evaluated based on mean squareerrors (MSE) and mean absolute percentage errors (MAPE).