Multivariate time series short term forecasting using cumulative data of coronavirus

Area of Science:

• Epidemiology
• Computational Biology
• Public Health

Background:

• Coronaviruses are highly contagious respiratory pathogens.
• Machine learning and time series analysis can analyze epidemic data for forecasting.
• Accurate forecasting models are crucial for combating infectious diseases.

Purpose of the Study:

• To perform short-term forecasting of cumulative COVID-19 incidences and mortality.
• To compare the accuracy of mathematical and deep learning models for epidemic forecasting.
• To analyze the impact of vaccination, temperature, and humidity on virus spread.

Main Methods:

• Utilized extended Susceptible-Exposed-Infected-Recovered (SEIR), Long Short-Term Memory (LSTM), and Vector Autoregression (VAR) models.
• Integrated hospitalization, mortality, vaccination, and quarantine data into the SEIR model.
• Employed Mean Absolute Error (MAE), Root Mean Square Error (RMSE), and Mean Absolute Percentage Error (MAPE) for evaluation.

Main Results:

• The Long Short-Term Memory (LSTM) deep learning model achieved the highest forecasting accuracy.
• Comparative analysis was conducted on eight severely affected nations.
• Vaccination impact, ambient temperature, and relative humidity effects on virus dissemination were explored.

Conclusions:

• Deep learning models, particularly LSTM, offer superior performance for short-term COVID-19 forecasting.
• Forecasting models incorporating diverse data improve prediction accuracy.
• Environmental factors and vaccination strategies significantly influence epidemic trajectories.