Computational graph-based framework for integrating econometric models and machine learning algorithms in emerging data-driven analytical environments

Abstract

In an era of big data and emergence of disrupting mobility technologies, statistical models have been utilized to uncover the influence of significant factors, and machine learning algorithms have been used to explore complex patterns in large datasets. Focusing on discrete choice modeling applications, this research aims to introduce computational graph (CG)-based frameworks for integrating the strengths of econometric models and machine learning algorithms. Specifically, multinomial logit (MNL), nested logit (NL), and integrated choice and latent variable (ICLV) models are selected to demonstrate the performance of the graph-oriented functional representation. Furthermore, the calculation of gradients in the log-likelihood function is accomplished using automatic differentiation (AD). Using the 2017 National Household Travel Survey data and synthetic datasets, we compare estimation results from the proposed methods with those obtained from Biogeme and Apollo. The results indicate that the CG-based choice modeling approach can produce consistent estimates of parameters with substantial computational efficiency.

KEYWORDS:

• Computational graphs (CGs)
• automatic differentiation (AD)
• multinomial logit (MNL)
• nested logit (NL)
• integrated choice and latent variable (ICLV)
• and gradient calculation

Acknowledgements

The authors gratefully acknowledge anonymous reviewers and Dr. Michel Bierlaire for providing constructive comments and informative suggestions. This research was supported by the Center for Teaching Old Models New Tricks (TOMNET) (grant number 69A3551747116), which is Tier 1 University Transportation Centers sponsored by the US Department of Transportation. The second author is partially supported by NSF grant number CMMI 1663657 ‘Real-time Management of Large Fleets of Self-Driving Vehicles Using Virtual Cyber Tracks’; The authors confirm contribution to the paper as follows; study conception and design: T. Kim, X. Zhou, R. Pendyala; data preparation: T. Kim; analysis and interpretation of results: T. Kim, X. Zhou, R. Pendyala; draft manuscript preparation: T. Kim, X. Zhou, R. Pendyala. All authors reviewed the results and approved the final version of the manuscript.

Disclosure statement

No potential conflict of interest was reported by the author(s).

Additional information

Funding

This research was supported by the Center for Teaching Old Models New Tricks (TOMNET) (grant number 69A3551747116), which is Tier 1 University Transportation Centers sponsored by the US Department of Transportation. The second author is partially supported by NSF grant number CMMI 1663657 ‘Real-time Management of Large Fleets of Self-Driving Vehicles Using Virtual Cyber Tracks’; Division of Civil, Mechanical and Manufacturing Innovation.