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

Urban rail transit (URT) systems operate in heterogenous environments where their performance is affected by many exogenous factors. However, conventional benchmarking methods assume homogeneity of many of these factors which could result in misleading comparisons of performance. This study provides a methodological contribution to the transit benchmarking literature through a systemic data-driven method which accommodates heterogeneity among URT. A unique international dataset of 36 URT systems in year 2016 is utilised. Operators are clustered based on indicators of operational performance through machine learning algorithms which enables like-for-like comparisons of performances. Data envelopment analysis with bootstrapping is then used to evaluate operators’ efficiency performance within a cluster. Further, ANOVA and post-hoc tests are applied to explore variations and correlations among different aspects of performance. Our clustering results corroborate the natural geographic grouping of the systems. Further, we highlight the complexity of the definition of service quality in the transit sector.