A fast shapelet discovery algorithm with symbolic fourier approximation
Cun Ji, Shijun Liu, Chenglei Yang, Li Pan, Lei Wu · 2016
Time series classification has been attracting great interest over the past decade.One of the most promising recent approaches is to find shapelets.A shapelet is one fragment of a time series that can be used to represent class characteristic of the time series.Classifier based on shapelets is interpretable, more accurate and faster.However, the time taken to find shapelet is enormous.For this, we propose a fast shapelet discovery algorithmwith symbolic Fourier approximation.In our algorithm, every time series is carried out by discrete Fourier transform to reduce the noise firstly.After that, we use symbols to approximate represent for the transformed data with the help of multiple coefficient binning.Finally, we use fast shapelet algorithm to get the best shapelet.Results fromexperiments show that our algorithm has a higher accuracy and interpretable classification results.