Estimating Express Train Preference of Urban Railway Passengers Based on Extreme Gradient Boosting (XGBoost) using Smart Card Data. Issue 11 (November 2021)

• Record Type: Journal Article
• Authors: Hak Lee, Eun; Kim, Kyoungtae; Kho, Seung-Young; Kim, Dong-Kyu; Cho, Shin-Hyung
• Is Part Of: Transportation research record. Volume 2675:Issue 11(2021)
• Page Start: 64
• Page End: 76
• Publication Date: 2021-11
• Subjects: Transportation -- Periodicals; Roads; Transport -- Périodiques; Routes -- Périodiques; Routes -- Conception et construction -- Périodiques; Roads; Transportation; 388.05
• DOI: 10.1177/03611981211013349
• Languages: English
• ISSNs: 0361-1981

Abstract

As the share of public transport increases, the express strategy of the urban railway is regarded as one of the solutions that allow the public transportation system to operate efficiently. It is crucial to express the urban railway's express strategy to balance a passenger load between the two types of trains, that is, local and express trains. This research aims to estimate passengers' preference between local and express trains based on a machine learning technique. Extreme gradient boosting (XGBoost) is trained to model express train preference using smart card and train log data. The passengers are categorized into four types according to their preference for the local and express trains. The smart card data and train log data of Metro Line 9 in Seoul are combined to generate the individual trip chain alternatives for each passenger. With the dataset, the train preference is estimated by XGBoost, and Shapley additive explanations (SHAP) is used to interpret and analyze the importance of individual features. The overall F1 score of the model is estimated to be 0.982. The results of feature analysis show that the total travel time of the local train feature is found to substantially affect the probability of express train preference with a 1.871 SHAP value. As a result, the probability of the express train preference increases with longer total travel time, shorter in-vehicle time, shorter waiting time, and few transfers on the passenger's route. The model shows notable performance in accuracy and provided an understanding of the estimation results.