Benchmarking the performance of urban rail transit systems: a machine learning application

Abstract

SHORT SUMMARY Urban rail transit systems operate in heterogenous environments. Distinguishing between inherent performance and the role of efficiencies due to differing environmental and system-specific characteristics is challenging. This study provides a data-driven benchmarking method which accommodates heterogeneity in operational performance among urban rail systems. Using an international dataset of 36 metros in year 2016, operators are clustered into peer groups through clustering algorithms based on operational characteristics. ANOVA and post-hoc tests are then applied to explore variations between clusters. Finally, efficiency performance benchmarking is conducted through Data Envelopment Analysis. Our clustering results corroborate to the natural geographic grouping of the systems. Moreover, our results show that the use of an aggregated index is inadequate to represent the operator’s overall quality-of-service. Finally, results show that clustering operators into groups based on similarities in their operational characteristics would introduce more meaningful benchmarks for best practices as they are more likely to be attainable.