Generative adversarial networks for biomedical time series forecasting and imputation.

Area of Science:

• Artificial Intelligence
• Machine Learning
• Biomedical Informatics

Background:

• Time series data are prevalent in many scientific fields, including biomedicine.
• Missing data and the need for future value prediction are common challenges.
• Generative Adversarial Networks (GANs) have emerged as a powerful tool for complex data generation and analysis.

Purpose of the Study:

• To systematically review and summarize research on Generative Adversarial Networks (GANs) for time series imputation and forecasting.
• To investigate the specific application and effectiveness of GANs within the biomedical domain.
• To analyze the characteristics and performance of GANs across various time series tasks.

Main Methods:

• A comprehensive literature search was conducted using PubMed, Web of Science, and Scopus.
• 1057 publications were initially identified, with 33 ultimately selected based on inclusion criteria.
• Eligible studies were categorized by GAN topology, loss functions, inputs, outputs, and application domains.

Main Results:

• Generative Adversarial Networks (GANs) demonstrate capability in both imputing missing time series values and forecasting future trends.
• The application of GANs spans diverse domains, with significant utilization in biomedical research.
• No specific GAN architecture or methodology was found to be universally superior for biomedical time series tasks.
• GANs, originally developed for image data, are successfully applied to non-visual time series data.

Conclusions:

• Generative Adversarial Networks (GANs) are versatile tools for time series imputation and forecasting across various scientific disciplines.
• While GANs are widely applied in biomedicine, current research does not indicate a single best-performing approach for this domain.
• The adaptability of GANs to non-visual time series data broadens their applicability in fields like bioinformatics and clinical data analysis.