Physics-guided framework for climate dependent disease prediction: coupling compartment model with deep learning.

Area of Science:

• Epidemiology
• Climate Science
• Artificial Intelligence

Background:

• Influenza-like illness (ILI) is a climate-sensitive disease requiring accurate prediction methods.
• Current models struggle to integrate climate data and capture complex transmission dynamics.
• Deep learning models often lack interpretability and overlook climate-disease interactions.

Purpose of the Study:

• To develop a physics-guided deep learning framework for enhanced ILI prediction.
• To integrate dynamical knowledge from compartmental models with multi-source data learning.
• To improve the accuracy and interpretability of ILI forecasting by incorporating climate factors.

Main Methods:

• Utilizing a physics-informed neural network (PINN) to embed differential equation constraints.
• Employing a physics-guided representation alignment mechanism to integrate PINN with LSTM.
• Training the model on ILI and climate data from Lanzhou and Xi'an, China.

Main Results:

• The proposed framework demonstrated superior prediction accuracy and trend consistency compared to five advanced baselines.
• Achieved significant reductions in prediction errors (NRMSE, SMAPE, ND, MAPE) in Lanzhou.
• Derived explicit functional relationships between transmission rates and climate factors using symbolic regression.

Conclusions:

• The physics-guided representation alignment mechanism enhances ILI prediction accuracy under varying climatic conditions.
• The framework provides valuable methodological support for epidemic prediction.
• Offers new insights into the complex interplay between climate and disease transmission.