UTA - NM: Explaining Stated Preferences with Additive Non-Monotonic Utility Functions

UTA methods use linear programming techniques for finding additive utility functions that best explain stated preferences. However, most UTA methods including the popular UTA-Star are limited to monotonic preferences. UTA-NM (Non Monotonic) is inspired by UTA Star but allows non-monotonic partial utility functions if they decrease total model error. The shape of the utility functions is determined automatically while overfitting is prevented by balancing the model error with model simplicity. The resulting program is linear and convex, but it requires significantly more CPU time than other UTA methods. The evaluation of the method on a synthetic task achieves the same Pearson Coefficient between the model and stated preferences as Despotis Non-Monotonic UTA. Unlike this method, UTA-NM does not require the provision of information on the shape of the utility function neither is it restricted to one change of shape per criterion.