Time Series K-Nearest Neighbors Classifier Based on Fast Dynamic Time Warping

Abstract

In the paper, a new Time Series classifier, which based on K-Nearest Neighbors (KNN) and Fast Dynamic Time Warping (FDTW), is presented. Fast dynamic time warping is particularly suitable for suitable for detecting signal similarity, which has an important character when we want to classify time series. K-Nearest Neighbors, which be used to slove regress and classify tasks, is a famous machine learning method. In this paper, we used FDTW as Features, and KNN as classifier. The algorithm forms a cluster, then comparing the characteristics of the signals to be classified. The time series of UCR was used in the experiment. By comparing classification results of fast dynamic time warping and neural networks, we can prove that the method is feasible, to a certain extent, improve the accuracy of signal classification.