Natural language, stocks, videos and almost any sensory measurement, including traffic data, have at least one thing in common: They are time series. Their future courses are determined by the (potentially very complex) interplay of their history, their inner structure and underlying rules as well as randomness and external factors. Both, automatic classification of time series (e.g. does this surveillance video contain violence?) and prediction their future development (e.g. language, stocks etc.) are of high interest. Deep Learning as a method originates from the computer vision domain, even so it recently achieved great successes across many domains, including time series domains. This master thesis explores the suitability of deep learning for time series classification and high-level prediction using a big real-world crowd sensed traffic data set in the context of electric mobility (DFKI project SADA). Goals are the classification of trips to users (who drives?), user types (young/old?), vehicles (model), trip types (highway/freeway) as well as the prediction of individual destinations or global events like jams.