Location based learning of user behavior for proactive recommender systems in car comfort functions

In-car comfort functions decrease driving stress and therefore increase safety. Drivers typically do not use all available comfort functions optimally, if at all, in every situation they would offer a substantial increase in comfort. Automating such functions as proactive recommender systems would exploit the full potential for decreasing driver stress. Because comfort functions are highly dependent on the driver’s habits, learning the individual user behavior is necessary. We propose a probabilistic method for modelling and predicting location dependent user behavior of comfort function activations. The model applies second-order uncertainty to evaluate the certainty about inferred parameter values and it deals with novelty and decaying observations explicitly. The results of this study show that the use of probabilistic models for learning location based user behavior in car comfort functions is a promising technique and gives reason to further investigate this area of studies.