Bus Passenger Demand Modelling Using Time-Series Techniques- Big Data Analytics

Background: Public transport demand forecasting is the fundamental process of transport planning activity. It plays a pivotal role in the decision making, policy formulations and urban transport planning procedures. In this paper, public bus passenger demand forecasting model is developed using a no...

Full description

Saved in:
Bibliographic Details
Published in:The open transportation journal 2019, Vol.13 (1), p.41-47
Main Authors: Cyril, Anila, Mulangi, Raviraj H., George, Varghese
Format: Article
Language:English
Citations: Items that this one cites
Items that cite this one
Online Access:Get full text
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:Background: Public transport demand forecasting is the fundamental process of transport planning activity. It plays a pivotal role in the decision making, policy formulations and urban transport planning procedures. In this paper, public bus passenger demand forecasting model is developed using a novel approach. The empirical passenger demand for a bus depot is modelled and forecasted using a data-driven method. The big data generated by Electronic Ticketing Machines (ETM) used for issuing tickets and collecting fares is sourced as the data for demand modelling. This big data is time indexed and hence has the potential for use in time-series applications which were not previously explored. Objectives: This paper studies the application of time-series method for forecasting public bus passenger demand using ETM based time-series data. The time-series approach used is the four Holt-Winters’ modeling methods. Holt-Winters’ additive and multiplicative models with and without damping have been empirically compared in this study using the data from the inter-zonal buses. The data used in the study is a part of the transaction on ticket sales by Kerala State Road Transport Corporation (KSRTC) maintained at the Trivandrum City depot of an Indian state Kerala, for the period between 2010 and 2013. The forecasting performance of four time-series models is compared using Mean Absolute Percentage Error (MAPE) and the model goodness of fit is determined using information criteria. Conclusion: The forecasts indicate that multiplicative models with and without damping, which better account for seasonal variations, outperform the additive models.
ISSN:1874-4478
1874-4478
DOI:10.2174/1874447801913010041