Prediction of COVID-19 confirmed cases combining deep learning methods and Bayesian optimization

Area of Science:

• Epidemiology
• Computational Science
• Data Science

Background:

• The COVID-19 pandemic has caused significant global health and economic disruptions.
• Accurate forecasting of COVID-19 cases is crucial for effective governmental intervention strategies.

Purpose of the Study:

• To develop and evaluate novel hybrid deep learning models for COVID-19 time series forecasting.
• To leverage Bayesian optimization for enhanced hyperparameter tuning and model performance.

Main Methods:

• Proposed three hybrid models combining multi-head attention, Long Short-Term Memory (LSTM), and Convolutional Neural Network (CNN) with Bayesian optimization.
• Implemented a multiple-output forecasting strategy for predicting multiple future time points.
• Utilized publicly available COVID-19 data from Johns Hopkins University's Coronavirus Resource Center.

Main Results:

• The hybrid deep learning models demonstrated superior performance compared to a benchmark model.
• Achieved a mean SMAPE of 0.25 for short-term forecasting (10 days ahead).
• Obtained a mean SMAPE of 2.59 for long-horizon forecasting.

Conclusions:

• Hybrid deep learning approaches integrating Bayesian optimization offer a powerful tool for accurate COVID-19 case forecasting.
• These advanced models provide reliable predictions for both immediate and extended future periods, aiding public health policy.