A Deep Learning Approach for Predicting Bus Passenger Demand Based on Weather Conditions

• Published in: Transport and telecommunication Vol. 21; no. 4; pp. 255 - 264
• Main Authors: Fontes, Tânia, Correia, Ricardo, Ribeiro, Joel, Borges, José Luís
• Format: Journal Article
• Language: English
• Published: Riga Sciendo 01-12-2020; De Gruyter Poland
• Subjects: Artificial neural networks; Bus passenger demand; Deep learning; Multilayer perceptrons; Predition; Regression analysis; Regression models; Ridership; Travel demand; Travel modes; Urban public transport; Weather; Weather conditions
• ISSN: 1407-6179; 1407-6160; 1407-6179
• DOI: 10.2478/ttj-2020-0020

This work apply a deep learning artificial neural network model – the Multilayer Perceptron – as a regression model to estimate the demand of bus passengers. Transit bus ridership and weather conditions were collected over a year from a medium-size European metropolitan area and linked under the assumption: individuals choose the travel mode based on the weather conditions that are observed during (a) the departure hour, (b) the hour before or (c) two hours prior to the travel start. The transit ridership data were also labelled according to the hour of the day, day of the week, month, and whether there was a strike and/or holiday or not. The results show that the prediction error of the model decrease by ~9% when the weather conditions observed two hours before travel start is taken into account. The model sensitivity analyses reveals that the worst performance is obtained for a strike day of a weekday in spring (typically Wednesdays or Thursdays).