Applying Dynamic Time Warping to Analyze Shape Similarities in Time Series Data Cover Image

Приложение на динамичното времево изкривяване за анализ на сходството във формата на времеви редове
Applying Dynamic Time Warping to Analyze Shape Similarities in Time Series Data

Author(s): Slaveya Zhelyazkova
Subject(s): Social Sciences, Economy, Education, Business Economy / Management, Higher Education
Published by: Икономически университет - Варна
Keywords: Dynamic time warping; Cluster analysis; dtwclust; Real house price indices
Summary/Abstract: The paper explores the potential of dynamic time warping (DTW) to detect shape-based similarities in time series data and assesses its effectiveness for cluster analysis using the "dtwclust" package in R. Theoretical considerations are demonstrated with an empirical analysis of real house price indices for 46 countries, utilizing OECD data spanning 2010–2023. Results indicate that employing DTW significantly influences clustering outcomes, as it enables the detection of shape similarities in time series, even in cases of temporal shifts and asynchronous patterns. This may lead to the identification of fewer clusters compared to standard clustering approaches. Furthermore, the choice of distance measure—Euclidean versus Manhattan—affects the clustering structure, with validity indices suggesting that DTW using Manhattan distance achieves better-defined, more distinct clusters. Comparative analysis with conventional hierarchical clustering highlights consistencies when similar distance metrics are applied.

  • Page Range: 133-145
  • Page Count: 13
  • Publication Year: 2024
  • Language: Bulgarian
Toggle Accessibility Mode