Data representation for time series data mining: time domain approaches
Focus Article
Published Online: Dec 25 2016
DOI: 10.1002/wics.1392
Can't access this content? Tell your librarian.
In most time series data mining, alternate forms of data representation or data preprocessing is required because of the unique characteristics of time series, such as high dimension (the number of data points), presence of random noise, and nonlinear relationship of the data elements. Therefore, any data representation method aims to achieve substantial data reduction to a manageable size, while preserving important characteristics of the original data, and robustness to random noise. Moreover, appropriate choice of a data representation method may result in meaningful data mining. Many high level representation methods of time series data are based on time domain approaches. These methods preprocess the original data in the time domain directly and are useful to understand the behavior of data over time. Piecewise approximation, data representation by identification important points, and symbolic representation are some of the main ideas of time domain approaches, and widely used in various fields. WIREs Comput Stat 2017, 9:e1392. doi: 10.1002/wics.1392 This article is categorized under: Statistical and Graphical Methods of Data Analysis > Dimension Reduction Statistical and Graphical Methods of Data Analysis > Data Reduction, Smoothing, and Filtering
Time series representation by piecewise aggregate approximation (PAA), perceptually important points (PIPs), and symbolic aggregate approximation (SAX). The dimension of the original data has been reduced from N = 200 to n = 10 by PAA and SAX, and to n = 11 by PIPs.
[
Normal View
|
Magnified View
]