Travel mode choice: a data fusion model using machine learning methods and evidence from travel diary survey data

In this paper, we present a series of machine learning approaches for better understanding people's travel mode choice. The widely used Logit model is dependent on the assumption that the utility items are independent, violating this assumption caused inconsistent parameter estimations and biased predictions. To improve the prediction accuracy of mode choice, this paper employs the data fusion model based on stacking strategy and proposes a hybrid model of the unsupervised Denoising Autoencoder (DAE) combining with the supervised Random Forest (RF). A variety of features that may impact mode choice behavior are ranked and selected by using the feature selection algorithms. The proposed model, which is particularly useful and powerful in the choice behavior analysis and outperforms other widely used classifiers, is verified by travel diary data from Germany and Switzerland. The results can be used for better understanding and effectively modeling of human travel mode choice behavior.