A novel ensemble ARIMA-LSTM approach for evaluating COVID-19 cases and future outbreak preparedness

Area of Science:

• Epidemiology
• Data Science
• Public Health

Background:

• The COVID-19 pandemic presents significant global public health and economic challenges.
• Time series analysis of COVID-19 confirmed cases is crucial for understanding disease spread.
• Data from India, Brazil, Russia, and the United States were analyzed.

Purpose of the Study:

• To develop a more accurate forecasting model for COVID-19 confirmed cases.
• To compare the performance of a hybrid ARIMA-LSTM model against established time series models.
• To enhance preparedness and response strategies for public health crises.

Main Methods:

• A hybrid model integrating Auto-Regressive Integrated Moving Average (ARIMA) and Long Short-Term Memory (LSTM) was developed.
• The hybrid model leverages ARIMA for linear trends and LSTM for nonlinear dependencies.
• Performance was evaluated against baseline models including ARIMA, Gated Recurrent Unit (GRU), LSTM, and Prophet.

Main Results:

• The hybrid ARIMA-LSTM model achieved a Mean Absolute Percentage Error (MAPE) of 2.4%, outperforming all benchmark models.
• Gated Recurrent Unit (GRU) showed the best performance among baseline models with a MAPE of 2.9%.
• LSTM achieved a MAPE of 3.6%, indicating lower accuracy compared to the hybrid approach.

Conclusions:

• The ARIMA-LSTM hybrid model demonstrates superior forecasting accuracy for COVID-19 cases compared to individual models and existing hybrid approaches.
• The model's effectiveness was validated across multiple countries using various accuracy metrics.
• Improved forecasting can lead to better public health preparedness, resource allocation, and intervention strategies.