Amplitude-Based Time Series Data Clustering Method

In the paper, we propose an amplitude-based time series data clustering method. When we analyze the trend index movement in economy, shape-based clustering does not work well because the standardization/z-normalization is required in advance on the input data and the standardization removes the amplitude/variance information from the original data. Then, the flat fluctuation may often become a large-variance fluctuation by the standardization, which is a problem. To solve the problem, we proposed a method by Amplitude-based time series data clustering method which uses Euclidean distance of Euclidean distances as the distance measurement. In the paper, we investigate the performance of the method, using the real stock prices data. The data are the indexed growth rate patterns of stock prices. Our proposed method could divide the companies’ stocks as we humans did, and the result met our requirements. The proposed amplitude-based time series data clustering method is helpful onomic indexed growth data clustering.