The domain of time series classification spans a wide range of data and applications (like language, stock market, EEG data, Videos and any kind of sensory recordings over time). Originating from the computer vision field, deep learning became increasingly successful and therefore popular across various domains, including language processing - a typical time series domain. In my master's thesis I investigate the suitability of deep learning for the classification of multivariate time series data by the example of a big crowd sensed automotive data set comprising longtime CAN-Bus recordings of a fleet of electrical cars. Goals are the classification of single trips to users (who drives?), user types (young/old?), vehicles (model), trip types (highway/freeway) as well as the prediction of destinations. The approach is compared to nearest neighbor classification (with dynamic time warping) as a classical machine learning method.
