A new approach to the space–time analysis of big data: application to subway traffic data in Seoul

Abstract

A prevalent type of big data is in the form of space–time measurements. Cyclostationary empirical orthogonal function (CSEOF) analysis is introduced as an efficient and valuable technique to interpret space–time structure of variability in a big dataset. CSEOF analysis is demonstrated to be a powerful tool in understanding the space–time structure of variability, when data exhibits periodic statistics in time. As an example, CSEOF analysis is applied to the hourly passenger traffic on Subway Line #2 of Seoul, South Korea during the period of 2010–2017. The first mode represents the weekly cycle of subway passengers and captures the majority (~97%) of the total variability. The corresponding loading vector exhibits a typical weekly pattern of subway passengers as a function of time and the locations of subway stations. The associated principal component time series shows that there are two occasions of significant reduction in the amplitude of the weekly activity in each year; these reductions are associated with two major holidays—lunar New Year and Fall Festival (called Chuseok in Korea). The second and third modes represent daily contrasts in a week and are associated with taking extra days off before or after holidays. The fourth mode exhibits an interesting upward trend, which represents a general decrease in the number of subway passengers during weekdays except for Wednesday and an increase over the weekends.