Using Smart Card Data to Quantify the Disruption Impact on Urban Metro Systems

• Activity Number: 618 - Machine Learning for Big Data
• Type: Contributed
• Date/Time: Thursday, August 1, 2019 : 8:30 AM to 10:20 AM
• Sponsor: Section on Statistical Learning and Data Science
• Author(s): Nan Zhang* and Daniel Graham and Jose M. Carbo and Daniel Hörcher
• Companies: Imperial College London and Imperial College London and Imperial College London and Imperial College London
• Keywords: Metro systems; smart card data; propensity score matching; disruption impact

Abstract

Disruptions occur regularly in urban metro systems, which generally stop normal train services for at least five minutes. This paper applies a propensity score matching method to quantify the causal impact of metro disruptions based on smart card data. The motivation for this approach is that the occurrence of disruptions is not random. Factors such as time of day, weather condition, real-time travel demand and engineering design can influence both the probability of a station being disrupted and the outcome of disruptions, therefore impact estimations based on the random occurrence assumption may be biased. Our empirical results show that, at the system level, metro disruptions do cause substantial increases in passenger exits and reduce average travel speed. The magnitude of these impacts varies from station to station with the largest effects on travel demand concentrated in Central London and the largest effects on travel speed dispersed in outer London areas. These results have important implications for metro operators in relation to optimal recovery plans from disruptions.