COVID-19 spread control policies based early dynamics forecasting using deep learning algorithm

Area of Science:

• Epidemiology
• Artificial Intelligence
• Public Health

Background:

• Global pandemics like COVID-19 significantly impact societies and economies.
• Accurate forecasting of disease spread is crucial for implementing effective control measures.
• Predictive models must account for various intervention strategies to guide policy decisions.

Purpose of the Study:

• To develop and validate a deep learning model for precise COVID-19 forecasting in South Korea.
• To assess the influence of public health interventions on disease spread prediction accuracy.
• To present an optimized, lightweight model for real-time epidemic monitoring.

Main Methods:

• Utilized a Stacked Bi-directional Long Short-Term Memory (Stacked Bi-LSTM) deep learning network.
• Incorporated fourteen parameters, including control policies (school/work closures, event cancellations), into the forecasting model.
• Compared Stacked Bi-LSTM performance against traditional time-series and standard LSTM models using MAE, MAPE, and RMSE metrics.

Main Results:

• The Stacked Bi-LSTM model demonstrated superior accuracy in forecasting COVID-19 cases compared to traditional methods.
• Analysis revealed the significant impact of control policies on prediction accuracy.
• Investigated the effect of activation function (ReLU vs. Tanh) on model performance, showing improved accuracy with ReLU.

Conclusions:

• The Stacked Bi-LSTM offers a robust and accurate approach for predicting COVID-19 spread, integrating policy impacts.
• Findings provide valuable insights for policymakers to optimize resource allocation and intervention strategies.
• The model's accuracy in forecasting cases, deaths, recoveries, and quarantines aids in proactive public health management.