Less is more: Selecting the right benchmarking set of data for time series classification