Home  | Publications | KBS+19

Clustering Trend Data Time-Series Through Segmentation of FFT-Decomposed Signal Constituents

MCML Authors

Link to Profile Thomas Seidl PI Matchmaking

Thomas Seidl

Prof. Dr.

Director

Database Systems, Data Mining and AI

Abstract

When we are given trend data for different keywords, scientists may want to cluster them in order to detect specific terms which exhibit a similar trending. For this purpose the periodic regression on each of the time-series can be performed. We ask in this work: What if we not simply cluster the regression models of each time-series, but the periodic signal constituents? The impact of such an approach is twofold: first we would see at a regression level how similar or dissimilar two time-series are regarding their periodic models, and secondly we would be able to see similarities based on single signal constituents between different time-series, containing the semantic that although time-series may be different on a regression level, they may be similar on an constituent level, reflecting other periodic influences. The results of this approach reveal commonalities between time series on a constituent level that are not visible in first place, by looking at their plain regression models.

inproceedings

LWDA 2019

Conference on Lernen. Wissen. Daten. Analysen. Berlin, Germany, Sep 30-Oct 02, 2019.

Authors

D. Kazempour • A. Beer • O. Schrüfer • T. Seidl

Links

PDF

Research Area

 A3 | Computational Models

BibTeXKey: KBS+19

Back to Top