Revisiting kernel logistic regression under the random utility models perspective. An interpretable machine-learning approach

The success of machine-learning methods is spreading their use to many different fields. This paper analyses one these methods, the Kernel Logistic Regression (KLR), from point view Random Utility Model (RUM) and proposes KLR specify utilities in RUM, freeing modeler need postulate a functional relation between features. A Monte Carlo simulation study conducted empirically compare with Multinomial Logit (MNL) method, Support Vector Machine (SVM) Forests (RF). We have shown that, using simulated data, only method that achieves maximum accuracy leads an unbiased willingness-to-pay estimator for non-linear phenomena. In real travel mode choice problem, RF achieved highest predictive accuracy, followed by KLR. However, allows calculation indicators such as value time, which great importance context transportation.