Time Series and Principal Component Analysis

Summarise with VIRAT
Authors: Anna Tscharre
Cited by: 0

Summary

No summary yet. VIRAT can read this paper and write a plain-language summary of its aim, methods, key findings and limitations.

Summarise with VIRAT Free to try with your Science Catalogs account.

Abstract

This thesis investigates how Principal Component Analysis (PCA) and related methods can be applied to time series, specifically to analyse periodic behaviour. After covering basic concepts in time series analysis and the classical PCA framework, we cover two PCA-based non-parametric techniques: Singular Spectrum Analysis (SSA) and frequency-domain PCA (sometimes called dynamic PCA). SSA is used to decompose both univariate and multivariate time series into interpretable components, in particular, into cycles of different period lengths. Frequency PCA extends PCA to the frequency domain, allowing the study of how variables co-vary at specific frequencies. While both SSA and frequency-domain PCA are discussed, the thesis places greater emphasis on SSA.SSA is applied to the historical sunspot number series, successfully recovering the well-known 11-year solar cycle as well as the longer Gleissberg cycle. We further use Multichannel SSA to analyse the same sunspot series complemented by Kp-index data, as well as a multivariate meteorological dataset. Frequency PCA is then used to extend the analysis of the meteorological data, evaluating periodic structures beyond the dominant annual cycle. Together, these results demonstrate that these non-parametric PCA-based approaches provide a flexible and powerful framework for uncovering periodicities and dynamics in both univariate and multivariate time series.

Fields of Study & Subjects

Time Series Analysis and Forecasting Statistical Methods and Applications Statistical and numerical algorithms

Publication details

Journal / source
University of Klagenfurt
Publication date
1 January 2026
Language
EN
Document type
Thesis