Machine Learning Travel Mode Choices: Comparing the Performance of an Extreme Gradient Boosting Model with a Multinomial Logit Model

Abstract

The multinomial logit (MNL) model and its variations have been dominating the travel mode choice modeling field for decades. Advantages of the MNL model include its elegant closed-form mathematical structure and its interpretable model estimation results based on random utility theory, while its main limitation is the strict statistical assumptions. Recent computational advancement has allowed easier application of machine learning models to travel behavior analysis, though research in this field is not thorough or conclusive. In this paper, we explore the application of the extreme gradient boosting (XGB) model to travel mode choice modeling and compare the result with an MNL model, using the Delaware Valley 2012 regional household travel survey data. The XGB model is an ensemble method based on the decision-tree algorithm and it has recently received a great deal of attention and use because of its high machine learning performance. The modeling and predicting results of the XGB model and the MNL model are compared by examining their multi-class predictive errors. We found that the XGB model has overall higher prediction accuracy than the MNL model especially when the dataset is not extremely unbalanced. The MNL model has great explanatory power and it also displays strong consistency between training and testing errors. Multiple trip characteristics, socio-demographic traits, and built-environment variables are found to be significantly associated with people’s mode choices in the region, but mode-specific travel time is found to be the most determinant factor for mode choice.

Analysis of Travel Mode Choice in Seoul Using an Interpretable Machine Learning Approach

“…Although travel cost is an important variable in the travel mode choice, the NHTS data used in this study did not include the respondents' travel cost such as fuel cost, parking cost, and transit fares. erefore, the effect of travel cost does not consider in the analysis like other studies using the NHTS data [1,7,8]. After a data-cleaning process, in which the trips were removed with very long activity duration and travel time, a total of 172,889 trips taken by 76,190 individuals were used.…”

“…e recent emergence of new travel modes such as ridesourcing, ride-hailing, and autonomous vehicles and the evolution of new mobility services such as mobility as a service and mobility on demand (known as MaaS and MoD, respectively) is changing travel behavior significantly [1].…”

“…It can represent complex relationships between mode choices and input variables in a data-driven manner rather than making strict assumptions about the data [7]. Many previous studies have reported the use of an ML approach to model travel mode choice [1,[6][7][8][9][10][11]. ese authors have generally reported improvements in the prediction performance of ML approaches compared to MNLbased models.…”

“…Several studies have performed additional analyses of the prediction results to complement the evaluation of performance. Wang and Ross proposed an extreme gradient boosting (XGB) model for predicting travel mode choice [1]. Using a relatively comprehensive dataset, the authors measured the relative importance of variables in the training process of the XGB and estimated the importance of correlated variables that cannot be explained using the MNL model.…”

A Machine Learning Approach to “Revisit” Specialization and Sampling in Institutionalized Practice

“…XGBoost belongs to the group of tree learning algorithms, whose impact has been widely recognized in a number of machine learning and challenges (Chen & Guestrin, 2016; for a comparison of different supervised machine learning methods compare, for example, Caruana & Niculescu-Mizil, 2006). Especially XGBoost recently enhanced great reputation by consistently winning competitions hosted by the machine learning competition site Kaggle (Chen & Guestrin, 2016;Wang & Ross, 2018). Although using these new approaches might require some reorientation in thinking, the clear evidence of their strong predictive performance and reliable identification of relevant variables and interactions (Elith et al, 2008) make them an extremely promising approach for the research problem at hand.…”