Incorporating weather conditions and travel history in estimating the alighting bus stops from smart card data. (February 2020)

• Record Type: Journal Article
• Authors: Tang, Tianli; Liu, Ronghui; Choudhury, Charisma
• Is Part Of: Sustainable cities and society. Volume 53(2020)
• Publication Date: 2020-02
• Subjects: Smart card data -- Machine learning -- Gradient boosting decision tree -- Alighting bus stop; Sustainable urban development -- Periodicals; Sustainable buildings -- Periodicals; Urban ecology (Sociology) -- Periodicals; 307.76
• DOI: 10.1016/j.scs.2019.101927
• Languages: English
• ISSNs: 2210-6707

Abstract

Highlights: New data driving framework for bus ridership estimation. Gradient boosting decision tree (GBDT) machine learning (ML) algorithm for estimation. GBDT outperforms traditional ML algorithms on estimation precision. Weather features further improves the precision of the estimation. Travel history enhances the recall ability of the model. Abstract: Origin-destination flow of passengers in bus networks is a crucial input to the public transport planning and operational decisions. Smart card systems in many cities, however, record only the bus boarding information (namely an open system), which makes it challenging to use smart card data for origin-destination estimations and subsequent analyses. This study addresses this research gap by proposing a machine learning approach and applying the gradient boosting decision tree (GBDT) algorithm to estimate the alighting stops of bus trips from open smart card data. It advances the state-of-the-art by including, for the first time, weather variables and travel history of individuals in the GBDT algorithm alongside the network characteristics. The method is applied to six-month smart card data from the City of Changsha, China, with more than 17 million trip-records from 700 thousand card users. The model prediction results show that, compared to classic machine learning methods, GBDT not only yields higher prediction accuracy but more importantly is also able to rank the influencing factors on bus ridership. The results demonstrate that incorporation of weather variables and travel history further improves the prediction capability of the models. The proposed GBDT-based framework is flexible and scalable: it can be readily trained with smart card data from other cities to be used for predicting bus origin-destination flow. The results can contribute to improved transport sustainability of a city by enabling smart bus planning and operational decisions.