Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Grasynda: Graph-based Synthetic Time Series Generation

Luis Amorim, Moises Santos, Paulo J. Azevedo, Carlos Soares, Vitor Cerqueira

TL;DR

Grasynda addresses data scarcity in time-series forecasting by introducing a graph-based synthetic data generator that converts a univariate series into a state-transition graph and uses the resulting transition matrix to sample realistic sequences. A quantile-graph variant and STL-based preprocessing for non-stationarity enhance robustness, enabling effective augmentation across diverse datasets. Extensive experiments on six datasets with three forecasting architectures show Gr беҙunda consistently competitive or superior performance compared with existing augmentation methods and outperforms TSMixup, underscoring the value of graph-based temporal representations for data generation. The approach is publicly available, highlighting practical impact for improving forecast accuracy in data-constrained settings.

Abstract

Data augmentation is a crucial tool in time series forecasting, especially for deep learning architectures that require a large training sample size to generalize effectively. However, extensive datasets are not always available in real-world scenarios. Although many data augmentation methods exist, their limitations include the use of transformations that do not adequately preserve data properties. This paper introduces Grasynda, a novel graph-based approach for synthetic time series generation that: (1) converts univariate time series into a network structure using a graph representation, where each state is a node and each transition is represented as a directed edge; and (2) encodes their temporal dynamics in a transition probability matrix. We performed an extensive evaluation of Grasynda as a data augmentation method for time series forecasting. We use three neural network variations on six benchmark datasets. The results indicate that Grasynda consistently outperforms other time series data augmentation methods, including ones used in state-of-the-art time series foundation models. The method and all experiments are publicly available.

Grasynda: Graph-based Synthetic Time Series Generation

TL;DR

Grasynda addresses data scarcity in time-series forecasting by introducing a graph-based synthetic data generator that converts a univariate series into a state-transition graph and uses the resulting transition matrix to sample realistic sequences. A quantile-graph variant and STL-based preprocessing for non-stationarity enhance robustness, enabling effective augmentation across diverse datasets. Extensive experiments on six datasets with three forecasting architectures show Gr беҙunda consistently competitive or superior performance compared with existing augmentation methods and outperforms TSMixup, underscoring the value of graph-based temporal representations for data generation. The approach is publicly available, highlighting practical impact for improving forecast accuracy in data-constrained settings.

Abstract

Data augmentation is a crucial tool in time series forecasting, especially for deep learning architectures that require a large training sample size to generalize effectively. However, extensive datasets are not always available in real-world scenarios. Although many data augmentation methods exist, their limitations include the use of transformations that do not adequately preserve data properties. This paper introduces Grasynda, a novel graph-based approach for synthetic time series generation that: (1) converts univariate time series into a network structure using a graph representation, where each state is a node and each transition is represented as a directed edge; and (2) encodes their temporal dynamics in a transition probability matrix. We performed an extensive evaluation of Grasynda as a data augmentation method for time series forecasting. We use three neural network variations on six benchmark datasets. The results indicate that Grasynda consistently outperforms other time series data augmentation methods, including ones used in state-of-the-art time series foundation models. The method and all experiments are publicly available.
Paper Structure (20 sections, 7 equations, 3 figures, 2 tables)

This paper contains 20 sections, 7 equations, 3 figures, 2 tables.

Table of Contents

  1. Introduction
  2. Background
  3. Time Series Forecasting
  4. Data Augmentation
  5. Time Series Analysis using Network Science
  6. The Grasynda Methodology
  7. Time Series Graph
  8. Discretization.
  9. Graph Construction.
  10. Data generation
  11. Quantile-based Grasynda
  12. Time series quantile graph
  13. Data preprocessing: handling non-stationarity
  14. Experiments
  15. Experimental Design
  16. ...and 5 more sections

Figures (3)

  • Figure 1: Example plot of a time series in the time domain (left); its network representation (middle); and the corresponding transition matrix (right).
  • Figure 2: Workflow behind Grasynda for building a graph from a time series and using it to create synthetic time series.
  • Figure 3: Example plot of a time series in the time domain (left); its network representation using a quantile graph (middle); and a synthetic time series generated from the graph using Grasynda (right).