ON THE USE OF MACHINE LEARNING TECHNIQUES AND DISCRETE CHOICE MODELS IN MODE CHOICE ANALYSIS

Background: The mode choice stage is a critical aspect that transportation experts rely on to develop a robust transportation system for a particular region. Various techniques are utilized to model mode choice behavior, including Discrete Choice Models (DCMs) and Machine Learning (ML) techniques. However, existing reviews typically focus on either DCMs or ML techniques, and reviews that cover both categories often concentrate on one category while merely mentioning some techniques from the other. This paper aims to address this gap by examining the principal DCMs and ML techniques published over the past four years, differentiating between models based on the granularity level, namely aggregate and disaggregate models. Additionally, a comprehensive discussion is conducted on the accuracy of the different models used in the reviewed articles.

Methods: This paper provides a thorough and enhanced analysis of travel mode choice models and analysis techniques used in articles published on "ScienceDirect" from 2020 to 2023. To ensure a comprehensive coverage of the subject, a meticulous search strategy was employed, utilizing targeted keywords. As a result, a total of 38 articles were carefully selected for detailed examination and analysis.

Results: The findings of this study highlight the suitability of different modeling approaches for varying levels of analysis. Discrete Choice Models demonstrate effectiveness in aggregate-level analyses, whereas Machine Learning Techniques prove more appropriate for disaggregate-level analyses. Moreover, the study suggests that employing hybrid models can potentially yield a promising solution to attain enhanced prediction accuracy without compromising interpretability.

Conclusions: The examination of selected articles revealed several key points. Firstly, there is a concentration of studies on travel mode choice in European countries, China, and the USA, indicating a need for more research in developing countries. Secondly, the reviewed articles often lack in-depth analysis of individual behavior and fail to consider external factors like weather or seasons when employing disaggregate models. Thus, future studies should leverage technological advancements and explore new factors influencing mode choice behavior. Additionally, there is a need for further research on hybrid models that combine Discrete Choice Models (DCMs) with Machine Learning (ML) techniques or deep learning approaches. This research can provide guidance for practitioners unfamiliar with these methods and aid in the design of effective transportation policies. Lastly, considering the variety of models available, it is crucial to understand the extent to which these models can be generalized to different contexts, emphasizing the importance of studying model applicability and generalizability in diverse settings.