[A comparative study of time series models in predicting COVID-19 cases]

Area of Science:

• Epidemiology
• Data Science
• Computational Biology

Background:

• Accurate forecasting of COVID-19 cases is crucial for public health response.
• Time series models offer potential for predicting disease trajectories.

Purpose of the Study:

• To evaluate and compare the predictive performance of Autoregressive Integrated Moving Average (ARIMA) and Recurrent Neural Network (RNN) models for COVID-19.
• To assess model performance across different countries with varying epidemiological characteristics.

Main Methods:

• Collected daily COVID-19 confirmed case data for the USA, India, and Brazil (April 1 - September 30, 2020).
• Developed and applied ARIMA and RNN models to predict case numbers.
• Utilized Mean Absolute Percentage Error (MAPE) and Root Mean Square Error (RMSE) for performance evaluation.

Main Results:

• ARIMA models showed MAPEs of 13.18% (USA), 9.18% (India), and 17.30% (Brazil).
• RNN models showed MAPEs of 15.27% (USA), 7.23% (India), and 26.02% (Brazil).
• RMSE values varied, with ARIMA generally outperforming RNN in the USA and Brazil, and RNN outperforming ARIMA in India.

Conclusions:

• Model performance for COVID-19 prediction is country-specific.
• The ARIMA model demonstrated superior predictive accuracy in the USA and Brazil.
• The RNN model proved more effective for forecasting COVID-19 cases in India.