A Deep Learning Approach to Infer Employment Status of Passengers by Using Smart Card Data

Abstract

Understanding the employment status of passengers in public transit systems is significant for transport operators in many real applications such as forecasting travel demand and providing personalized transportation service. This paper develops a deep learning approach to infer a passenger's employment status by using smart card data (SCD) with a household survey. This paper first extracts an individual passenger's weekly travel patterns in different travel modes from the raw SCD as a three-dimensional image. A deep learning architecture, called a thresholding multi-channel convolutional neural network, was developed to predict an individual's employment status. The approach proposed here solves two critical problems of using the SCD for employment status studies. First, it automatically incorporates learning temporal features in different travel modes without the need for handcrafted travel feature design. Second, it considers the class-imbalance problem by leveraging the ensemble of oversampling and thresholding techniques. By applying our approach to a real dataset collected from the metropolitan area of London, U.K., about 72% of passengers were correctly categorized into six types of employment statuses. The promising results show the tight correlation between temporal travel behavior, mode choice, and social-demographic roles. To the best of our knowledge, this is the first paper to infer employment status by using the SCD.Index Terms-Deep learning, employment status inference, travel mode choice, smart card data, temporal travel behavior.

Semantic enrichment of secondary activities using smart card data and point of interests: a case study in London

“…(Bagchi and White 2005). Notwithstanding the wide range of positive characteristics, smart card data present several challenges such as: estimating a commuter's destination if public transport does not ask for alighting information (Gordon et al 2013), making demographic predictions if socio-demographic information is not accessible due to privacy concerns (Zhang, Cheng, and Sari Aslam 2019;Zhang and Cheng 2020), detecting activities in order to estimate a trip's purpose by linking smart card data with auxiliary data sources, such as land use maps and POIs (Devillaine, Munizaga, and Trépanier 2012; Kuhlman 2015; Sari Aslam and Cheng 2018;Yang et al 2019).…”

A graph deep learning method for short‐term traffic forecasting on large road networks

“…Thus, latent features are automatically extracted to model underlying, complicated, and nonlinear relationships in the data. In transportation research area, DL has been successfully applied to tackle many problems (Hashemi & Abdelghany, ; Nabian and Meidani, ; Zhang & Cheng, ). Concerning traffic flow prediction, Lv et al.…”

Exploring the Relationship Between Travel Pattern and Social-Demographics Using Smart Card Data and Household Survey

“…However, it lacks the social-demographic information of passengers to further explore 'who are the card carriers', 'why they behaved differently' and 'what factors affect their behaviours', which are crucial to better understand the users' travel demand and mobility patterns. Fortunately, leveraging household survey data, it might further explore the relationship between human travel patterns and their socialdemographic roles (Zhang et al 2018;Zhang et al 2019), which can help operators make better transportation planning and provide passengers with more personalised services.…”