Improving influenza surveillance based on multi-granularity deep spatiotemporal neural network.

Area of Science:

• Epidemiology
• Computer Science
• Public Health

Background:

• Influenza poses a significant public health threat, necessitating accurate risk prediction for effective management.
• Traditional prediction models often overlook crucial spatiotemporal dynamics, limiting their accuracy.
• Spatiotemporal data, reflecting regional and temporal interactions, is key to understanding influenza transmission patterns.

Purpose of the Study:

• To develop an advanced deep neural network for accurate influenza risk prediction.
• To address the limitations of existing models by incorporating spatiotemporal feature mining.
• To improve public health management through enhanced influenza forecasting.

Main Methods:

• Proposed a novel end-to-end spatiotemporal deep neural network.
• Employed a two-stream architecture combining convolutional and recurrent neural networks for feature extraction.
• Utilized a dynamically parametric-based fusion method to integrate spatiotemporal features for prediction.

Main Results:

• The proposed model demonstrated superior performance across all evaluation metrics on two influenza-like illness (ILI) datasets (US-HHS and SZ-HIC).
• Achieved state-of-the-art results in influenza risk prediction compared to established forecasting models.
• Validated the model's effectiveness in capturing complex spatiotemporal dependencies.

Conclusions:

• The developed spatiotemporal deep neural network offers a powerful tool for accurate influenza risk prediction.
• Integrating spatiotemporal features significantly enhances forecasting capabilities for respiratory diseases.
• This approach holds promise for improving public health surveillance and intervention strategies.