Pattern frequency representation for time series classification
Sergey Milanov, Olga Georgieva · 2016
The paper presents a new method for data transformation. The obtained data format enhances the efficiency of the time series classification. The transformation is realized in four steps, namely: time series segmentation, segment's feature representation, segments' binning and pattern frequency representation. The method is independent of the nature of the data, and it works well with temporal data from diverse areas. Its ability is proved by classification of different real datasets and is also compared with the results of other classification methods.