Output details
11 - Computer Science and Informatics
University of East Anglia
Classification of time series by shapelet transformation
<15>Shapelets are discriminatory time series subsequences that represent a new type of feature for time series classification. Shapelets are ideal for problems where class membership is defined by localised, phase-independent shapes embedded in longer series. Through rigorous experimentation, we demonstrate that by separating the shapelet transformation from the classification stage we can get significantly better accuracy than the standard shapelet algorithm. Shapelets have been shown to be effective in areas such as motion and image outline classification. This paper extends two papers presented at KDD and SDM 2012, both of which have approximately 10% acceptance rate.