The Study on Algorithms for Piecewise Linear Representation of Time Series

摘要

It is a key point for efficient performances of data mining algorithms to express time series availably. After the study and analysis of three algorithms, i. e. sliding window, top-down and bottom-up, we present an enhanced PLR algorithm based on bottom-up and dual threshold value. The concept of relative point mean error ( RPME ) is proposed and used to measure the linear degree of sub-series, which produced by bottom-up algorithm. This method improves the accuracy of piecewise linear representation. An example on supermarket selling data is demonstrated.