ResLogit: A residual neural network logit model for data-driven choice modelling

This paper presents a novel deep learning-based travel behaviour choice model. Our proposed Residual Logit (ResLogit) model formulation seamlessly integrates Deep Neural Network (DNN) architecture into multinomial logit Recently, DNN models such as the Multi-layer Perceptron (MLP) and Recurrent (RNN) have shown remarkable success in modelling complex noisy behavioural data. However, econometric studies argued that machine learning techniques are ‘black-box’ difficult to interpret for use analysis. We develop data-driven extends systematic utility function incorporate non-linear cross-effects using series of residual layers skipped connections handle identifiability estimating large number parameters. The structure accounts heterogeneity arising from substitution, interactions with non-chosen alternatives other effects manner. describe formulation, estimation, interpretability examine relative performance implications our present an illustrative example on classic red/blue bus scenario example. For real-world application, we mode dataset analyze characteristics compared traditional neural networks formulations. findings show ResLogit approach significantly outperforms MLP while providing similar Multinomial