Predicting Demands of COVID-19 Prevention and Control Materials via Co-Evolutionary Transfer Learning

Area of Science:

• Epidemiology
• Machine Learning
• Public Health

Background:

• The COVID-19 pandemic has strained medical resources, necessitating accurate demand prediction for critical supplies.
• Estimating medical material demands is complex due to intricate relationships with infection rates and limited historical data.

Purpose of the Study:

• To propose a novel Co-evolutionary Transfer Learning (CETL) method for accurately predicting medical material demands during the COVID-19 pandemic.
• To address the challenge of insufficient historical data and complex demand-infection rate relationships.

Main Methods:

• Developed a CETL method utilizing fuzzy deep contractive autoencoders (CAEs) for each prediction task.
• Employed intrapopulation evolution for task-specific knowledge and interpopulation evolution for shared knowledge across domains.
• Transferred knowledge from related epidemics (SARS, bird flu) and disasters.

Main Results:

• CETL demonstrated high prediction accuracy for medical material demands during COVID-19.
• The method outperformed existing state-of-the-art transfer learning and multitask learning models.
• Effective knowledge transfer was achieved even with limited and imbalanced data from related tasks.

Conclusions:

• The proposed CETL method offers a robust solution for predicting medical supply demands during health crises.
• Leveraging diverse data sources through transfer learning enhances prediction accuracy and resource management.
• CETL provides a valuable tool for pandemic preparedness and response efforts.