Time series mining based on multilayer piecewise aggregate approximation

Time series is a ubiquitous data existed in different domains including finance, medicine, business and other industrial fields. Recently, time series data mining attracts much attention. In this paper, we propose multilayer piecewise aggregate approximation (MPA) to measure the Similarity of time series. The proposed method is constituted of two parts: multi-level segment method based on extreme value is used to extract important identification sub-series of time series. And piecewise aggregate approximation is used to transform the data and to extract features from time series so as to reduce data dimension. After that, dynamic time warping is applied to measure the distance between two time series. The experimental results demonstrate that the proposed method can extract features and reduce data dimension efficiently, with improving the efficiency and the accuracy of time warping distance method significantly.