Benchmarking the performance of mass transit systems via panel statistical modelling and machine learning techniques
File(s)
Author(s)
Awad, Farah
Type
Thesis
Abstract
Performance measurement and benchmarking has been successfully applied in many sectors of service provision and has become a powerful management tool. The concept of transit performance measurement is defined as a multi-dimensional process. However, in practice, the literature is segmented between different domains which results in a partial view of the overall performance. Mass transit systems have a sociotechnical nature. Their service domains are complex and interdependent and operate in heterogenous environments. Their performance is affected by a range of internal and external influencing factors – of both the supply service level and the demand for that service. However, current benchmarking practices of performance tend to overlook the interdependent nature of the relationship between service domains, and the contextual influences. Therefore, they are likely to provide misleading comparisons of performance which result in unrealistic setting of best practice.
This thesis focuses on the expansion of the multi-dimensional concept of transit performance benchmarking within a socio-technical context. Interactions among different domains of performance are explored and integrated to provide a holistic performance perspective. Furthermore, the practice of performance benchmarking is improved by reducing the external influences and effects of the operating environment on measures of relative performance.
The first part of this thesis aims to improve the performance benchmarking practices of urban rail transit by providing more contextualized comparisons of performance. This is achieved by applying machine learning algorithms to cluster operators into peer groups based on similarities in operational performance. Variations between clusters are explored and cluster profiles are created based on financial, operational, and environmental factors which helps in identifying patterns and correlations among different domains of performance. The second part of this research aims to examine the relationship between these two major aspects of performance in urban rail transit systems. This is achieved using panel data econometric models which control for endogeneity bias stemming from unobserved heterogeneity and reverse causality. A tractable framework which integrates cost, supply, demand, and quality of service is proposed. Finally, the third part of this thesis employs Artificial Neural Networks to develop predictive models of urban rail safety performance. An improved framework of safety performance benchmarking is proposed which accommodates heterogeneity in network and operational characteristics.
This thesis contributes to the mass transit performance literature by integrating multiple performance domains in benchmarking applications to give a holistic perspective of performance. Additionally, contextualized benchmarks are proposed using advanced panel statistical models and machine learning models that accommodate heterogeneity and non-linear interactions among performance domains and operating features. A unique dataset of over urban rail 30 systems around the world is used to generate robust estimates.
This thesis focuses on the expansion of the multi-dimensional concept of transit performance benchmarking within a socio-technical context. Interactions among different domains of performance are explored and integrated to provide a holistic performance perspective. Furthermore, the practice of performance benchmarking is improved by reducing the external influences and effects of the operating environment on measures of relative performance.
The first part of this thesis aims to improve the performance benchmarking practices of urban rail transit by providing more contextualized comparisons of performance. This is achieved by applying machine learning algorithms to cluster operators into peer groups based on similarities in operational performance. Variations between clusters are explored and cluster profiles are created based on financial, operational, and environmental factors which helps in identifying patterns and correlations among different domains of performance. The second part of this research aims to examine the relationship between these two major aspects of performance in urban rail transit systems. This is achieved using panel data econometric models which control for endogeneity bias stemming from unobserved heterogeneity and reverse causality. A tractable framework which integrates cost, supply, demand, and quality of service is proposed. Finally, the third part of this thesis employs Artificial Neural Networks to develop predictive models of urban rail safety performance. An improved framework of safety performance benchmarking is proposed which accommodates heterogeneity in network and operational characteristics.
This thesis contributes to the mass transit performance literature by integrating multiple performance domains in benchmarking applications to give a holistic perspective of performance. Additionally, contextualized benchmarks are proposed using advanced panel statistical models and machine learning models that accommodate heterogeneity and non-linear interactions among performance domains and operating features. A unique dataset of over urban rail 30 systems around the world is used to generate robust estimates.
Version
Open Access
Date Issued
2022-11-29
Date Awarded
01/05/2023
License URL
Advisor
Graham, Daniel J.
Publisher Department
Civil and Environmental Engineering
Publisher Institution
Imperial College London
Qualification Level
Doctoral
Qualification Name
Doctor of Philosophy (PhD)