The Nearest Neighbor Classifiers for Time Series with Complex Shape Features

Time series are widely used in various fields of daily life, and they usually have rich local shape features. However, most current similarity measures only consider the functional distance between points, ignoring the shape characteristics of the time series. To this end, this paper first proposes a time series local shape similarity measurement, which can effectively represent and discover the local shape features of time series. Secondly, this shape similarity is combined with existing distance measures to improve the classification accuracy of the algorithm. Finally, the experimental results of 40 UCR datasets verify the validity and interpretability of the proposed methods.