Predicting Short-term Bus Ridership with Trip Planner Data: A Machine Learning Approach
Z. Wang (TU Delft - Civil Engineering & Geosciences)
A.J. Pel – Mentor (TU Delft - Transport and Planning)
Trivik Verma – Graduation committee member (TU Delft - Policy Analysis)
P.K. Krishnakumari – Graduation committee member (TU Delft - Transport and Planning)
Niels Oort – Graduation committee member (TU Delft - Transport and Planning)
P. van Brakel – Graduation committee member (REISinformatiemroep B.V.)
More Info
expand_more
Other than for strictly personal use, it is not permitted to download, forward or distribute the text or part of it, without the consent of the author(s) and/or copyright holder(s), unless the work is under an open content license such as Creative Commons.
Abstract
To address the increasing passenger demand in the coming years and make public transport less crowded and delayed, insights into predicted passenger flow are needed. A wide range of studies has used and validated that smart card data can be one of the sound bases for predicting short-term passenger demand. However, it also has several disadvantages, such as the relatively long collection time, the insufficiency to reflect the relationship between passenger behavior and ridership. Trip planner data, which emerged as a type of real-time transit information, could reduce the perceived waiting time of passengers and increase the transit ridership due to the improved satisfaction. Combining these two types of data could potentially cater to the interest of operators in matching the vehicle supply and passenger flow demand at an operational level. Our results show that it is novel and useful to incorporate trip planner data in short-term ridership prediction, however, entirely based on this kind of data would be inaccurate. Random Forest Regression outperforms the other six models that we have selected. The request-related features (variables) can take up 20% of the importance of short-term ridership prediction.