Enhancing destination choice modelling with neural network techniques

The destination choice model treats individuals as the unit of analysis, providing flexibility in examining factors influencing destination choice behaviour. However, conventional linear-in-parameters utility functions may fail to capture complex and nonlinear relationships between explanatory variables and destination utility, potentially biasing estimated variable sensitivities. To address this limitation, this study introduces a neural network (NN)-based approach into the destination choice modelling framework to learn nonlinear functional forms from data and flexibly approximate destination utility. The proposed approach is applied to non-commuters’ shopping trips using data from the 2019 Shanghai Household Travel Survey. Partial Dependence Plots (PDPs) are further employed to interpret the effects of explanatory variables. The results show that the NN-based model achieves a higher goodness-of-fit than the conventional model and more flexibly captures variable effects on destination utility. The findings demonstrate the potential of NN-based destination choice modelling for travel demand analysis and transportation planning.