Understanding Commuter Mode Choice in Rome: A Comparative Analysis of Neural Networks and Multinomial Logit for Sustainable Mobility

Understanding how commuters select their travel modes is vital for advancing sustainable urban mobility and reducing reliance on private vehicles. This study compares the performance of Multinomial Logit (MNL) and Neural Network (NN) models using travel survey data collected from employees in Rome, Italy. Both models are evaluated under a consistent preprocessing and validation pipeline to ensure fair comparison. While the NN model achieves slightly higher predictive accuracy, a paired t-test confirms that the performance difference is not statistically significant. Elasticity analysis and SHAP-based interpretation are employed to assess the impact of key variables. SHAP results show that both models highlight factors such as public transport inefficiency, cost, and parking difficulty. The MNL model offers sharper prioritization of key features, while the NN reveals broader, non-linear patterns. These findings support hybrid approaches that combine MNL interpretability with NN flexibility to improve behavioral insights and guide sustainable transportation policies.