ALGORITHMIC METHODS FOR SEGMENTATION OF TIME SERIES: AN OVERVIEW

  • Miodrag Lovrić Full time Professor, Faculty of Economics, University of Kragujevac and Visiting Professor at Federal University of Pernambuco, Brazil
  • Marina Milanović Teaching Assistant, Faculty of Economics, University of Kragujevac
  • Milan Stamenković Teaching Assistant, Faculty of Economics, University of Kragujevac
Keywords: time series, data mining, segmentation, piecewise linear approximation, algorithm, approximation error

Abstract

Adaptive and innovative application of classical data mining principles and techniques in time series analysis has resulted in development of a concept known as time series data mining. Since the time series are present in all areas of business and scientific research, attractiveness of mining of time series datasets should not be seen only in the context of the research challenges in the scientific community, but also in terms of usefulness of the research results, as a support to the process of business decision-making. A fundamental component in the mining process of time series data is time series segmentation. As a data mining research problem, segmentation is focused on the discovery of rules in movements of observed phenomena in a form of interpretable, novel, and useful temporal patterns. In this Paper, a comprehensive review of the conceptual determinations, including the elements of comparative analysis, of the most commonly used algorithms for segmentation of time series, is being considered.

Downloads

Download data is not yet available.
Published
2019-03-19