Travel mode choice modelling plays a critical role in predicting passengers’ travel demand and planning local transportation systems. Researchers commonly adopt classical Random Utility Models to analyse individual decision-making based on the utility theory. Recently, with an increasing interest in applying Machine Learning techniques, a number of studies have used these methods for modelling travel mode preferences for their excellent predictive power. However, none of these studies proposes machine learning models that investigate the regional heterogeneity of travel behaviours. To address this gap, this study develops a Random Effect-Bayesian Neural Network (RE-BNN) framework to predict and explain travel mode choice across multiple regions by combining the Random Effect (RE) model and the Bayesian Neural Networks (BNN). The results show that this model outperforms the plain Deep Neural Network (DNN) regarding prediction accuracy and is more robust across different datasets. In addition, in terms of interpretation, the capability of RE-BNN to learn the travel behaviours across regions has been demonstrated by offset utilities, choice probability functions and local travel mode shares.

出行方式选择建模在预测乘客出行需求和规划当地交通系统方面发挥着关键作用。研究人员通常采用经典的随机效用模型来分析基于效用理论的个人决策。最近，随着对应用机器学习技术的兴趣日益浓厚，许多研究已经使用这些方法对出行方式偏好进行建模，因为它们具有出色的预测能力。然而，这些研究都没有提出研究旅行行为的区域异质性的机器学习模型。为了解决这一差距，本研究开发了一个随机效应-贝叶斯神经网络 (RE-BNN) 框架，通过结合随机效应 (RE) 模型和贝叶斯神经网络 (BNN) 来预测和解释跨多个区域的出行方式选择。结果表明，该模型在预测精度方面优于普通深度神经网络 (DNN)，并且在不同数据集上更稳健。此外，在解释方面，RE-BNN 学习跨区域出行行为的能力已通过偏移效用、选择概率函数和本地出行模式份额得到证明。