STELAR: Spatio-temporal Tensor Factorization with Latent Epidemiological Regularization.

Area of Science:

• Epidemiology and Public Health
• Computational Science and Data Analytics

Background:

• Accurate prediction of epidemic disease transmission, such as COVID-19, is vital for effective public health interventions.
• Existing methods often struggle with simultaneous multi-region forecasting and long-term trend prediction.

Approach:

• Developed a novel tensor method, STELAR (Spatio-Temporal tensor factorization with latent Epidemiological model Regularization).
• Constructed a 3-way spatio-temporal tensor (location, attribute, time) of case counts.
• Incorporated latent temporal regularization using discrete-time difference equations from epidemiological models for enhanced prediction.

Key Points:

• STELAR enables long-term, multi-region prediction by leveraging latent epidemiological dynamics, capturing common epidemic profile sub-types.
• The model improves collaborative learning and prediction accuracy by analyzing shared patterns across regions.
• Experiments with COVID-19 data (county- and state-level) demonstrate STELAR's ability to identify latent epidemic patterns.

Conclusions:

• STELAR significantly outperforms baseline methods in epidemic trend prediction.
• Achieved up to 21% lower root mean square error and 25% lower mean absolute error for county-level COVID-19 prediction.
• The method offers a powerful tool for public health officials to anticipate and manage disease outbreaks.