Visibility graph for time series prediction and image classification: a review.

Area of Science:

• Complex Systems Analysis
• Data Science
• Network Science

Background:

• Time series and image analysis are crucial across various scientific domains.
• Traditional methods include AI, physical models, and hybrid approaches.
• Complex networks offer a powerful framework for modeling complex systems.

Purpose of the Study:

• To review visibility graph algorithms for mapping time series and images to complex networks.
• To explore how these networks reveal properties of time series and images.
• To evaluate the efficacy of visibility graph algorithms in prediction and classification tasks.

Main Methods:

• Development and review of various visibility graph algorithms (e.g., horizontal visibility graph, image visibility graph).
• Application of statistical physics to analyze network topology and information.
• Utilizing local random walk algorithms and information fusion for time series forecasting.
• Employing machine learning models (SVM, LDA) for image classification based on network features.

Main Results:

• Visibility graph algorithms construct diverse network types (weighted, directed, multi-layered).
• Network analysis reveals underlying properties of time series and images.
• Proposed forecasting frameworks and classification methods demonstrate effectiveness.
• Simulations show visibility graph algorithms outperform existing methods in time series prediction and image classification.

Conclusions:

• Visibility graph algorithms establish a strong link between complex networks, time series, and image analysis.
• Complex networks serve as a vital tool for understanding data characteristics.
• The study highlights the potential and future directions for visibility graph applications.