A novel approach for COVID-19 Infection forecasting based on multi-source deep transfer learning.

Area of Science:

• Epidemiology
• Machine Learning
• Data Science

Background:

• Accurate prediction of COVID-19 infections is crucial for pandemic management.
• Data scarcity in specific regions poses a significant challenge for traditional machine learning models, leading to low prediction accuracy.
• Existing models often struggle with the high bias and low variance issues inherent in small datasets.

Purpose of the Study:

• To propose a novel Multi-Source Deep Transfer Learning (MSDTL) technique for accurate COVID-19 infection forecasting in data-scarce provinces.
• To enhance prediction accuracy by integrating population density and Gross Domestic Product (GDP) as key features.
• To leverage the Long Short-Term Memory (LSTM) recurrent neural network architecture for time-series prediction.

Main Methods:

• Developed an MSDTL approach combining deep learning (LSTM) with multi-source transfer learning.
• Incorporated provincial population density and GDP data alongside historical COVID-19 infection, mortality, and recovery rates.
• Evaluated the model's performance on a COVID-19 dataset from sixty-two provinces globally using Mean Absolute Percentage Error (MAPE) and Coefficient of Determination (R²).

Main Results:

• The proposed MSDTL approach significantly improved the accuracy of future COVID-19 infection predictions compared to models without transfer learning.
• The model demonstrated substantial enhancements, achieving up to a 96% improvement over its non-transfer learning counterpart.
• Experimental validation confirmed the importance of selected features (population density, GDP) for accurate forecasting.

Conclusions:

• MSDTL, particularly with LSTM and relevant socio-economic features, is highly effective for forecasting infectious diseases in regions with limited data.
• The integration of population density and GDP enhances predictive precision, offering a more robust tool for pandemic response.
• This methodology provides a scalable and accurate solution for epidemiological forecasting, crucial for efficient public health interventions.