Detecting Change Points in Time Series with Gaussian Processes
Keywords
Loading...
Authors
Issue Date
2024-08-31
Language
en
Document type
Journal Title
Journal ISSN
Volume Title
Publisher
Title
ISSN
Volume
Issue
Startpage
Endpage
DOI
Abstract
Time series data are prevalent in many fields of study, such as environmental sciences,
medical monitoring and finance. These datasets are observations of a system over time,
which can undergo changes at any given moment, and provide information about e.g.
how the system reacts after an intervention or transitions through different phases. This
study investigates the use of Gaussian processes to detect points where such changes
occur and to model time series data using Sequential Monte Carlo posterior estimation.
We compare the performance of our method to the Infinite Hidden Markov Model and
demonstrate similar or better performance at change point detection and fitting of the
data on various types of toy and real datasets. Furthermore, our method allows us
to classify different types of change points by using different types of kernels in the
Gaussian process. The most crucial limitation to our method is the computational cost,
which limits the size of usable datasets
Description
Citation
Supervisor
Faculty
Faculteit der Sociale Wetenschappen