Application of radial basis function and generalized regression neural networks in non-linear utility function specification for travel mode choice modelling

Application of soft computational methods, especially artificial neural networks, in examining individual traveller behaviour is not encountered frequently. In most of the relevant cited papers, the feed-forward back propagation neural network (FFBPNN) models or hybrid models of FFBPNNs are proposed. However the feed-forward back propagation algorithm has some drawbacks, which can easily lead the model to develop in an inaccurate direction. Throughout this study, two different algorithms, radial basis function neural network (RBFNN) and generalized regression neural network (GRNN), are employed to propose a new calibration process for travel mode choice analysis in a transportation modelling framework. The neural network methods are not applied directly to calibrate models but are used as a sub-process for alternative non-linear model specification on utility function. Results show both the surpassing of RBFNNs and GRNNs over frequently used FFBPNNs, and the superiority of neural network methods over a conventional statistical model, multivariate linear regression, during mode choice calibrations. Also having experienced the existence of a claim that ANNs can tackle the problem of travel choice modelling as well as, if not better than, the discrete choice approach [D.A. Hensher, T.T. Ton, A comparison of the predictive potential of artificial neural networks and nested logit models for commuter mode choice, Transp. Res., Part E Logist. Trans. Rev. 36 (3) (2000) 155-172], use of such soft computing tools in studying traveller behaviour should be an autonomous part of a calibration process.