An enhanced temporal decision tree is described that allows a temporal feature to be tested at a node not merely at a given time instant but, rather, over a time window. Further, the recorded temporal data can be multi-source, with features recorded from multiple sources; multi-resolution, through subsampling; multi-scale, based on varying time windows; and multi-domain, through sampling in the frequency domain, for example. By allowing features to be tested over a time window, enhanced temporal decision trees can be used to classify any temporal data or time series.