Functional approximation for the classification of smooth time series
Friedrich Melchert, Udo Seiffert, Michael L. Biehl · University of Groningen research database (University of Groningen / Centre for Information Technology) · 2016
Time series data are frequently analysed or classied by con- sidering sequences of observations directly as high-dimensional feature vectors. The presence of several hundreds or thousands of input dimen- sions can lead to practical problems. Moreover, standard algorithms are not readily applicable when the time series data is non-equidistant or the sampling rate is non-uniform. We present an approach that allows for a massive reduction of input dimensions and explicitly takes advan- tage of the functional nature of the data. Furthermore, the application of standard classication algorithms becomes possible for inhomogeneously sampled time series. The presented approach is evaluated by applying it to four publicly available time series datasets.