Predicting COVID-19 cases using bidirectional LSTM on multivariate time series

Area of Science:

• Epidemiology
• Data Science
• Public Health

Background:

• Accurate forecasting of COVID-19 spread is crucial for effective policy decisions.
• Existing forecasting methods may not fully capture complex influencing factors.

Purpose of the Study:

• To develop and validate a novel deep learning approach for forecasting cumulative COVID-19 cases.
• To enhance prediction accuracy by incorporating country-specific similarities and lockdown data.

Main Methods:

• Utilized a bidirectional Long Short-Term Memory (Bi-LSTM) network for multivariate time series forecasting.
• Employed K-means clustering to group countries based on demographic, socioeconomic, and health indicators.
• Integrated lockdown measure data with cumulative case data for model training.

Main Results:

• The proposed Bi-LSTM model demonstrated superior performance in forecasting COVID-19 cases.
• Validation using Qatar's outbreak data from December 2020 showed the model's effectiveness.
• Quantitative evaluation confirmed outperformance against state-of-the-art forecasting techniques.

Conclusions:

• The integrated deep learning approach offers a more accurate method for COVID-19 propagation forecasting.
• Clustering countries and incorporating policy data improves the reliability of epidemiological predictions.
• This technique can aid policymakers in making informed decisions to control the pandemic.