Time Series Clustering Based on Dynamic Time Warping

Abstract

In general, solving prediction problems requires a series of operations for the data set such as preprocessing, partitioning, and structuring features, so as to fit a better prediction model. For time series data, it is divided into different data sets according to certain rules to achieve the effect of improving the accuracy of the prediction model. This paper proposes a more novel clustering method which the traditional Euclidean distance and dynamic time planning are separately weighted and combined to do the distance calculation method in clustering. A time series contains both a time dimension and a spatial dimension. Euclidean distance is mainly used for spatial distance calculation. Dynamic time warping can calculate the similarity calculation in time dimension, similar to the distance calculation in the spatial dimension. The measure of similarity of time series is a measure of the degree of similarity between two time series. It is verified by experiments that under the same prediction model, this novel clustering method is better than the Euclidean distance clustering method and the traditional dynamic time warping method.