Transfer Learning for COVID-19 cases and deaths forecast using LSTM network

Area of Science:

• Epidemiology
• Computer Science
• Machine Learning

Background:

• The COVID-19 pandemic necessitated accurate forecasting models for public health response.
• Existing epidemiological models faced challenges in rapidly predicting disease spread across diverse regions.

Purpose of the Study:

• To apply Transfer Learning in Long Short-Term Memory (LSTM) networks for forecasting new COVID-19 cases and deaths.
• To evaluate the efficacy of models trained on early-pandemic data for predicting disease spread in previously unaffected countries.

Main Methods:

• Utilized Transfer Learning techniques to adapt LSTM models trained on data from Italy and the United States.
• Performed single-step and multi-step forecasting of COVID-19 incidence and mortality.
• Validated model performance using data from Germany, France, Brazil, India, and Nepal.

Main Results:

• The Transfer Learning-based LSTM models demonstrated promising accuracy in forecasting COVID-19 trends.
• The method showed validity when applied to diverse geographical locations with varying infection dynamics.
• Forecasts provided valuable insights into potential disease trajectories.

Conclusions:

• Transfer Learning in LSTM networks offers a robust approach for predicting COVID-19 spread.
• The findings support the use of these models for informing public health policies and resource allocation.
• This methodology can aid policymakers in managing the ongoing threat of COVID-19.