Detecting Home and Work Locations Using Multiday Transit Smart Card Data: Comparison of Three Methods

Abstract

The identification of commuters’ home and work locations is crucial for urban and transport planning because it enables a better understanding of the urban spatial structure and commuting flow. Various methods have been developed for home and job location identification; however, the accuracy, reliability, and sensitivity of these methods have not been thoroughly examined. This study aimed to compare three commonly used approaches—the staying time method, the trip frequency method, and the hidden Markov chain method (HMM)—in terms of their adaptability and sensitivity to different scales of data, advantages and disadvantages by using smart card data of Beijing in 5 weeks of 2016. Our results showed significant differences among the three methods in identifying actual commuters. The staying time method had the largest error, while HMM was more intelligent in the recognition result due to its combination with historical inbound and outbound passenger flow rules. Although the staying time method was simple and easy to implement, it was unable to fully reflect the data’s characteristics. For larger amounts of sample data, the trip frequency method demonstrated faster processing efficiency; however, missing data had a significant impact on the results. Finally, the machine learning method was able to identify locations more intelligently than the other two approaches, although its algorithm’s time complexity and resource consumption were very high. These findings provided new insights into the application of big data in urban spatial research and offered suggestions for selecting the most appropriate identification method based on data and scenarios.