An algorithmic approach for collaborative-based prediction of user contexts in ubiquitous environments under consideration of legal implications

Context prediction is used to proactively adapt, e.g. services in a ubiquitous environment to users’ needs. Due to the fact that context prediction enables proactiveness, the significance for ubiquitous computing systems is high. To the best of our knowledge, wellknown approaches in context prediction only focus on a user’s history as a database whose next contexts have to be predicted. In case a user suddenly changes her behaviour in an unexpected way and does not follow her routine anymore, the context history of the user does not contain appropriate context information to provide reliable context predictions. Hence, context prediction algorithms that only rely on the user’s context history whose context has to be predicted, might fail. To overcome the gap of missing context information in a user’s context history, the Collaborative Context Prediction (CCP) approach is proposed. CCP takes advantage of existing direct and indirect relations, which may exist among the context histories of various users and therefore provides the possibility to forecast a user’s next context even if the user suddenly changes her expected routine. CCP is based on the Higher-order Singular Value Decomposition, which has already successfully been applied in existing recommendation systems. To provide an evaluation of CCP, it is assessed in three different experiments. In these experiments, results are carried out with respect to prediction accuracy. These results are compared to the results received by three state of the art context prediction approaches: the Alignment predictor, the StatePredictor and the ActiveLeZi prediction approach. In all three experiments, collaborative data sets are used as a basis for evaluation. Moreover, CCP is applied to a realistic collaborative use case, the proactive protection of pedestrians. CCP is used to proactively detect pedestrians that might be at risk to collide with a car nearby, using real movement data, measured by smartphones the pedestrians carried in their trouser pocket. Due to the fact that context prediction approaches primarily use personal contexts such as location data or users’ behaviour patterns, legal evaluation criteria are derived considering the principles of a user’s right to informational self-determination. Based on the derived legal evaluation criteria, the CCP approach and the state of the art context prediction approaches are examined. The evaluation results outline the compatibility of different context prediction approaches to a user’s right to informational self-determination. Finally, an approach for distributed and collaborative context prediction is presented in this thesis. This approach presents a possibility to overcome the identified legal problems caused by context prediction, especially by collaborative-based context prediction.