Horizontal Visibility Graph-Based Time Series Forecasting
SEFBIS Journal · 21.05.2026 · (1) · 1
DOI: 10.14267/sefbis.2026.01 ↗Kurz gefasst
Fachartikel in der Zeitschrift SEFBIS Journal, online erschienen am 21.05.2026. Verfasst von Ferenc Farkas. Die Arbeit nennt 14 Literaturangaben.
Abstract (Originalsprache)
Time series forecasting remains challenging when the underlying process is irregular, nonlinear, or structurally changing. This study presents a time series forecasting approach based on Horizontal Visibility Graphs (HVG), where sliding-window segments of a time series are transformed into graph structures and structural graph metrics are used as input features for forecasting models. The framework combines window-based segmentation, HVG construction, graph feature extraction, and supervised learning. The empirical analysis is conducted on a controlled subset of daily time series from the M4 dataset and compares HVG-based models with classical statistical benchmarks, namely ARIMA and Holt–Winters. The results show that HVG-derived structural metrics provide an informative feature space and can achieve competitive performance, although effectiveness is model-dependent. In a representative example, several HVG-based models outperform the statistical baselines. The findings indicate that HVG representations can serve as a complementary feature space for time series forecasting.
Zitieren
APA
Farkas, F. (2026). Horizontal Visibility Graph-Based Time Series Forecasting. SEFBIS Journal, 1. https://doi.org/10.14267/sefbis.2026.01
Vancouver
Farkas F. Horizontal Visibility Graph-Based Time Series Forecasting. SEFBIS Journal. 2026:1. doi:10.14267/sefbis.2026.01
BibTeX
@article{Farkas2026,
author = {Farkas, Ferenc},
title = {Horizontal Visibility Graph-Based Time Series Forecasting},
journal = {SEFBIS Journal},
year = {2026},
number = {1},
pages = {1},
doi = {10.14267/sefbis.2026.01}
}
EUVIMED