Foodborne Disease Risk Prediction Using Multigraph Structural Long Short-term Memory Networks: Algorithm Design and...

Area of Science:

• Epidemiology
• Public Health
• Data Science

Background:

• Foodborne diseases pose a significant global health threat, causing millions of deaths annually.
• Accurate prediction of foodborne disease risk is vital for effective public health management and intervention strategies.

Purpose of the Study:

• To develop a spatial-temporal risk prediction model for forecasting foodborne disease risks across diverse regions.
• To provide a data-driven tool to guide the prevention and control of foodborne diseases.

Main Methods:

• Designed an end-to-end framework utilizing a multigraph structural long short-term memory neural network with an encoder-decoder architecture.
• Incorporated spatial correlations through administrative area division and constructed adjacent graphs based on proximity, data similarity, regional function, and food exposure.
• Integrated attention mechanisms in spatial and temporal dimensions, along with external factors, to enhance prediction accuracy.

Main Results:

• The model achieved high F1 scores for single-month forecasts in various Chinese provinces (e.g., Beijing 0.822, Zhejiang 0.679).
• The model's highest F1 score surpassed existing state-of-the-art models by 20%, demonstrating superior predictive performance.
• Experimental validation using a long-term real-world dataset (2015-2019) confirmed the model's effectiveness.

Conclusions:

• The developed spatial-temporal risk prediction model effectively captures the spatio-temporal characteristics of foodborne disease data.
• The model accurately predicts future disease risks, offering valuable support for foodborne disease prevention and risk assessment strategies.