Modeling transportation mode choice through artificial neural networks

We aim at showing that artificial neural networks (ANN) can be an effective tool for travel demand analysis. Exiting literature show that ANN's can outperform commonly adopted models, derived from random utility theory, but they are based on hypothesis on user behavior, thus their parameters cannot clearly be interpreted. A new architecture, which one extra layer for perceived utility, will be analyzed to address those main drawbacks. This way an explicit utility function is introduced allowing to an interpretation of input variables as well as elasticity analysis