Novel cost-effective method for forecasting COVID-19 and hospital occupancy using deep learning.

Area of Science:

• Epidemiology
• Artificial Intelligence
• Health Informatics

Background:

• The COVID-19 pandemic severely strained global healthcare systems, increasing mortality and highlighting the need for predictive tools.
• Effective healthcare management during crises requires accurate forecasting of disease spread and resource needs.

Purpose of the Study:

• To develop and implement an accurate, low-cost predictive system for forecasting pandemic evolution, including COVID-19 cases and hospital occupancy.
• To improve decision-making in healthcare management through reliable, long-term predictions.

Main Methods:

• Utilized interconnected Long Short-Term Memory (LSTM) with double bidirectional LSTM (BiLSTM) layers.
• Implemented a novel preprocessing technique based on future time windows.
• Trained the model on 40% of the data, achieving accurate long-term predictions.

Main Results:

• The predictive system demonstrated accuracy in forecasting COVID-19 cases and hospital occupancy.
• Achieved improved Mean Absolute Error (MAE) < 161, Root Mean Square Error (RMSE) < 405, and Mean Absolute Percentage Error (MAPE) > 0.20.
• Enabled daily querying of cases for the next three days with minimal data.

Conclusions:

• The developed LSTM-BiLSTM model provides a powerful tool for healthcare systems, offering valuable insights for strategic planning.
• The system supports optimized allocation of healthcare and economic resources, improving public health outcomes.
• This approach offers a cost-effective solution for accurate, long-term pandemic forecasting.