<p>Travel behavior modeling handles heterogeneous data but standard models often yield overconfident predictions. We propose a trusted framework (E-TNN/EL-TNN) that explicitly distinguishes between first-order choice randomness and second-order uncertainty. By embedding categorical variables and fusing sources via Dempster-Shafer theory, mode probabilities are modeled as Dirichlet distributions to produce principled uncertainty mass. Experiments across four international datasets demonstrate superior accuracy and the lowest calibration error (ECE) among all baselines. Crucially, the framework maintains behavioral validity, with estimated parameters and Value of Time (VoT) consistent with economic theory. Stress tests confirm the model’s risk-awareness, where uncertainty appropriately increases with data noise level. By bridging predictive power with behavioral rigor, our framework provides a robust and transparent decision-support tool for transportation planning and policy analysis.</p> Graphical abstract <p></p>