Identifying commuters based on random forest of smartcard data. Issue 4 (6th March 2020)

• Record Type: Journal Article
• Authors: Mei, Zhenyu; Ding, Wenchao; Feng, Chi; Shen, Liting
• Is Part Of: IET intelligent transport systems. Volume 14:Issue 4(2020)
• Page Start: 207
• Page End: 212
• Publication Date: 2020-03-06
• Subjects: transportation -- traffic engineering computing -- smart cards -- data mining -- learning (artificial intelligence) -- public transport; smartcard data -- commuter flow -- metro passenger flow -- efficient method -- spatiotemporal commuting patterns -- Metro Line 2 -- one‐week transit smart card data -- spatiotemporal regularity -- individual commuters -- travel time -- travelling days -- transit commuters -- random forest algorithm -- high‐efficiency analysis method -- unique tag information -- questionnaire survey data -- validated random forest model -- metro commuters -- commuter traffic -- personal‐level commute models; Intelligent transportation systems -- Periodicals; Electronics in transportation -- Periodicals; 388.31205
• DOI: 10.1049/iet-its.2019.0414
• Languages: English
• ISSNs: 1751-956X

Abstract

Abstract : Commuter flow is an important part of metro passenger flow. The aim of this study is to develop an efficient and effective method to identify the spatiotemporal commuting patterns of Metro Line 2 in Hangzhou. Using one‐week transit smart card data and a questionnaire survey of Metro Line 2, the authors distinguished the spatiotemporal regularity of individual commuters, including first travel time, last travel time, and the number of travelling days on weekdays. This data could be used to identify transit commuters by leveraging ensemble learning approaches. The random forest algorithm was adopted as a low‐cost, high‐efficiency analysis method, and the classification model was established with the information of travel time, days of travelling, and the unique tag information in the questionnaire survey data. Then, numerical tests were carried out to show that the Precision and Recall rates of the proposed model could reach as high as 0.96 and 0.92, respectively. Finally, the validated random forest model was applied to identify metro commuters from the smartcard data. The results show that less than one‐third of passengers are commuter traffic and are mainly concentrated during peak hours. These extracted personal‐level commute models can be used as valuable information for the design and optimisation of public transportation networks.