Time series classification using visibility graph and persistent homology

View at publisher Summarise with VIRAT
Authors: Deanne Erica Valderama, Marc Aaron F Julian, Mark Lexter D. De Lara
Cited by: 0
References: 11

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

Time series classification often relies on machine learning methods where the quality of features significantly impacts classification accuracy. This study introduces an innovative approach to assess the impact of employing visibility graph features, enhanced by persistence diagrams, on time series classification. Specifically, we transform time series data into visibility graphs, extract graph-based features, and further enhance the feature set by incorporating persistence diagrams. We assess the performance of the Persistent Homology Classification Algorithm (PHCA), which uses persistent homology to estimate the topological features of a dataset that persists across multiple resolutions, against traditional classifiers such as Random Forest (RF), Logistic Regression (LR), and Support Vector Machine (SVM) across five real-world time series datasets. The empirical results indicate that PHCA outperforms other classifiers on two datasets, achieving the highest accuracy of 75.24% and 74.03% on the Plane and WormsTwoClass datasets, respectively. In comparison, Random Forest achieved 79.00% accuracy on the Trace dataset, and SVM reached 73.20% accuracy on the Computers dataset. Logistic Regression demonstrated balanced performance, though it did not exceed other classifiers on any specific dataset. These findings highlight that combining visibility graph features with persistence diagrams can enhance classification accuracy across diverse datasets, suggesting improvements in fields such as healthcare, finance, and other time-sensitive domains. This study advances time series analysis by introducing topological data analysis methods, providing an approach to feature engineering for classification tasks in varied applications.

Fields of Study & Subjects

Topological and Geometric Data Analysis Time Series Analysis and Forecasting Complex Systems and Time Series Analysis

Publication details

Journal / source
AIP conference proceedings
Volume
3462
Pages
140004–140004
Publication date
1 January 2026
Language
EN