Working with sensor-data creates tremendous amounts of data in short time. It's not surprising, that a single computer or even server can't handle this amount of data using traditional methods. The focus of my research lies on the efficient parallelization of operations on big data, working on databases themselves. Special attention will be drawn to methods of detection and prediction for activity in the field of assisting systems. Another aspect is the provenance management of data to support the error prune development and maintenance of prediction models. The last aspect of my research deals with the (semi-)automatic transformation of machine-learning-programs (e.g. R) into database-programs. This will hopefully allow the user to embed the new database-algorithms without changing anything in his machine-learning-program code.