Time series forecasting of Covid-19 using deep learning models: India-USA comparative case study

Area of Science:

• Epidemiology
• Data Science
• Public Health

Background:

• COVID-19's global impact necessitates accurate forecasting for mitigation.
• Understanding transmission dynamics is crucial for controlling the pandemic.

Purpose of the Study:

• To predict future COVID-19 conditions using deep learning.
• To compare the performance of different recurrent neural network variants for forecasting.

Main Methods:

• Comparative analysis of deep learning models including Stacked LSTM, Bi-directional LSTM, and Convolutional LSTM.
• Forecasting confirmed and death cases of COVID-19 for India and USA one month ahead.
• Utilizing datasets of confirmed and death cases for model training and validation.

Main Results:

• Convolutional LSTM achieved higher accuracy and lower error rates compared to Stacked LSTM and Bi-directional LSTM.
• The model successfully forecasted upward/downward trends in COVID-19 cases for both countries.
• Graphical visualization of forecasted trends aids in understanding disease progression.

Conclusions:

• Deep learning, specifically Convolutional LSTM, offers a robust approach for COVID-19 case prediction.
• Accurate forecasting can significantly assist researchers and policymakers in managing the pandemic.
• The findings support data-driven strategies to mitigate mortality and morbidity.