Short & long term forecasting of multimodal transport passenger flows with machine learning methods

Abstract

Estimating and forecasting travel demand is one of the major applications for smart card data analysis. Forecasting can be useful in order to plan both services and trips. Depending on the considered time horizon, prediction can output average travel demand in the long term while short-term forecasting can be useful in order to match transport supply to real-time demand. This paper investigates the problem of passenger flow forecasting in multimodal transport and considers train stations and bus and tram stops. The aim is to be able to predict the number of passengers entering each station or boarding at each stop. Both long- and short-term forecasting models are developed using machine learning models such as Random Forests (RF), Long-Short Term Memory (LSTM) neural networks as well as calendar models. Forecasting is performed either for the next year (in the case of long-term models) or for the next 15 minutes for train stations and tram stops and within the next 30 minutes for buses, which could be helpful for both passengers and operators. The experiments are carried out on a real 2 year smart card dataset provided by the Transport organization authority of Ile-de-France. We focus on 145 stations and stops located in the district of La Defense, which is a well-known major business district in Paris Metropolitan Area. Our results have proved the effectiveness of the forecasting approaches using the available data and machine learning models.