Ensembles for multivariate time series classification

Ruiz, Alejandro Pasos (2023) Ensembles for multivariate time series classification. Doctoral thesis, University of East Anglia.

[thumbnail of APR 240128 Final PhD Thesis.pdf]
Download (8MB) | Preview


Time Series Classification (TSC) involves learning predictive models for a discrete target variable from ordered, real-valued, attributes. Over recent years,
a new set of TSC algorithms have been developed which have significantly improved the previous state of the art. The main focus has been on univariate TSC, i.e. the problem where each case has a single series and a class label. In reality, it is more common to encounter multivariate TSC (MTSC) problems where multiple series are associated with a single label. Despite this, much less consideration has been given to MTSC than the univariate case. Therefore, this work focuses on MTSC from different perspectives. First, by introducing a set of 33 problems for MTSC in different areas called the UEA MTSC archive.
Second, by introducing the state-of-the-art algorithms and comparing on those problems. That experimentation concluded that HIVE-COTE2 (HC2) is the
current state of the art. Third, because of that, the remainder of this work focused on two ways to improve HC2: a) By improving one of the components (Shapelet Transform Classifier) and b) by Adding a preprocessing phase for dimension selection in order improve HC2 by removing the dimensions that do not contribute. In the first case, we were able to improve HC2 significantly for MTSC problems, and in the second case, there was no significant improvement in accuracy. Still, there were gains in decreasing the number of dimensions required and hence the run time.

Item Type: Thesis (Doctoral)
Faculty \ School: Faculty of Science > School of Computing Sciences
Depositing User: Nicola Veasy
Date Deposited: 15 Apr 2024 09:56
Last Modified: 15 Apr 2024 10:00
URI: https://ueaeprints.uea.ac.uk/id/eprint/94905


Downloads per month over past year

Actions (login required)

View Item View Item