Travel behaviour research in the age of machine learning: opportunities and challenges

New sources of large-scale mobility data have led to a growing interest in the application of Machine Learning (ML) models in travel behaviour studies. ML models, which are data-driven and have flexible forms, are especially powerful in capturing the high levels of nonlinearity in the data. This makes them well-suited for inferring useful modelling inputs (e.g. travel mode, purpose), from large-scale mobility data. Further, there is a growing interest to use ML for predicting travel choices (e.g., vehicle ownership, mode and destination choices) - either as a stand-alone tool or in conjunction with traditional models based on theories of economics. However, explaining, or interpreting ML algorithms is often a challenging task. Also, ML models do not directly provide the welfare measures required for making transport investments and intervention decisions. By presenting an overview of the applications of ML in travel behaviour research and discussing the opportunities they provide, this chapter attempts to bridge ML and traditional travel behaviour models. Further, the challenges of using ML models in travel behaviour studies are discussed and candidate methods for better interpreting the ML model outputs are explored.