Dynamic targets - adapting supervised learning to time series classification

To train a classifier with supervised learning appropriate targets have to be provided. In the case of time series classification this can be complicated if there is only one target for the whole time series, but the learning algorithm needs a target at each time step. In this paper a new technique is introduced, which is able to provide appropriate targets at each time step. This allows the use of more complex learning algorithms, which results in faster learning and better generalization