Modeling Individual Driving Route Preferences from Relative Feedback

A common, yet diicult, task for people is planning satisfactory driving routes. Automatic systems for driving directions exist, but one of their major diiculties is the range of individual diierence regarding what constitutes a good route and a bad route. We introduce an adaptive user interface for route planning that uses relative feedback to model individual route preferences. Our experiments test three adaptive algorithms for relative preferences on two feature sets. The results show that all algorithms perform similarly on the training data, but the simpler algorithms have the better test performance, probably because the more complex algorithms t the noise in the training data. We intend to improve the accuracy of our adaptive algorithms by feature engineering and incorporating domain knowledge about route preferences.