Comparing and Contrasting Choice Model and Machine Learning Techniques in the Context of Vehicle Ownership Decisions - Empirical comparison of modeling approaches

Aug 5, 2026

Upload by Himbo

Recent planning practice has increasingly considered Machine Learning (ML) techniques as an alternative to discrete choice models (CM), despite concerns about limited grounding in economic theory and reduced transferability. This study tests two hypotheses by modeling vehicle ownership choices using household survey data from Dhaka, Bangladesh in 2004, 2010, and 2019. CM implemented via multinomial logit is compared against neural networks and gradient boosting trees using log-likelihood and market-share error, showing the strongest overall results for a tailored MNL specification.

LanguageEnglish

CategoryResearch & Report

FormatPDF

Pages20

Reading Time50 min

AccessFree to download

Introduction

Background: CM vs ML in travel behavior modeling

Vehicle ownership decision modeling setup

Transportation Research Part A 173 (2023) 103727 Contents lists available at ScienceDirect Transportation Research Part A journal [homepage:](homepage: www.elsevier.com/locate/tra)[ www.elsevier.com/locate/tra](homepage: www.elsevier.com/locate/tra) | Comparing and contrasting choice model and machine learning techniques in the context of vehicle ownership decisions | | | | | | | --- | --- | --- | --- | --- | --- | | Azam Ali , Arash Kalatian , Charisma
Institute for Transport Studies, University of Leeds, UK | | | F. Choudhury | * | | | A R T I C L E I N F O | | A B S T R A C T | | | | | Keywords:
Choice modelling
Machine learning Explainable machine learning Vehicle ownership model Developing countries Car ownership | | In recent years, planners have started considering Machine Learning (ML) techniques as an alternative to discrete choice models (CM). ML techniques are primarily data-driven and typically achieve better prediction accuracy compared to CM. However, it is hypothesized that since the ML techniques do not have the strong grounding to economic theory as the CMs, they may not perform well in contexts that are radically different from the ‘training’ scenario. It is also hypothesized that the relative prediction performance may be affected by the metrics used for comparing the models.
This research aims to test these two hypotheses empirically by modelling vehicle ownership choices using household survey data from Dhaka, Bangladesh collected in 2004, 2010 and 2019. The performances of CM (multinomial logit) and ML techniques (neural networks and gradient boosting trees) have been compared using log-likelihood and mean absolute percentage error of market shares.
The results indicate that the multinomial logit model (MNL) with a piecewise linear transformation of the household income, has the best performance in terms of log-likelihood and mean absolute percentage error of market shares. This is followed by Neural Networks (NN) and Gradient Boosting Trees (GBT). The results thus provide empirical evidence that the ML techniques do not consistently outperform CM. Moreover, the difference in the performance of the models further increases if the prediction scenario is substantially different. This reinforces the hypothesis that CMs, with their behavioural underpinning, are better suited for long-term forecasting than data-driven ML approaches, especially if the population and network attributes are expected to change substantially. These findings will be useful for planners and policy makers in the selection of the appropriate tool for forecasting travel demand. | | | | 1. Introduction Travel behaviour models have historically relied on Choice Models (CM) based on theories of economics and psychology. However, the availability of large datasets on human mobility in recent years has led to an increased interest in deploying Machine Learning (ML) techniques to predict travel behaviour. These ML techniques use parametric and nonparametric algorithms to ‘learn’ directly from the data (Van Cranenburgh et al., 2022; Walker et al., 2019). Such data-driven learning enables researchers to model large and complex datasets without making explicit assumptions about the behavioural motivations or the relationship between the dependent and the * Corresponding author. E-mail addresses: [ts19aa@leeds.ac.uk](ts19aa@leeds.ac.uk) (A. Ali), [arash.kalatian@gmail.com](arash.kalatian@gmail.com) (A. Kalatian), [C.F.Choudhury@leeds.ac.uk](C.F.Choudhury@leeds.ac.uk) (C.F. Choudhury). [https://doi.org/10.1016/j.tra.2023.103727](https://doi.org/10.1016/j.tra.2023.103727) Available online 6 June 2023 0965-8564/© 2023 The Author(s). Published by Elsevier Ltd. This is an open access article under the CC BY license ([http://creativecommons.org/licenses/by/4.0/](http://creativecommons.org/licenses/by/4.0/)). independent variables. However, the lack of behavioural underpinning also poses the risk of

FAQs

What problem does the study address about CM and ML for vehicle ownership?

It evaluates whether ML techniques can outperform discrete choice models for vehicle ownership decisions, and whether performance degrades when the prediction scenario differs from training conditions.

Which datasets and time points are used for the empirical test?

The study uses household survey data from Dhaka, Bangladesh collected in 2004, 2010, and 2019.

How do the compared models perform and what do the results imply?

A multinomial logit model with a piecewise linear transformation of household income performs best on both log-likelihood and market-share error; neural networks and gradient boosting follow, indicating ML does not consistently beat CM and performs worse under substantially different scenarios.

Aug 5, 2026

Upload by Himbo