Improving performance of deep learning predictive models for COVID-19 by incorporating environmental parameters

Area of Science:

• Epidemiology and Public Health
• Data Science and Machine Learning
• Environmental Science

Background:

• The COVID-19 pandemic's global economic impact necessitates accurate forecasting for healthcare preparedness.
• Existing forecasting models often lack consideration of environmental influences on disease transmission.
• Effective prediction tools are crucial for managing healthcare resources and controlling pandemic spread.

Purpose of the Study:

• To analyze the impact of incorporating environmental parameters into forecasting models for daily COVID-19 cases.
• To compare the performance of univariate and multivariate Long Short-Term Memory (LSTM) models in predicting COVID-19 spread.
• To investigate the relationship between environmental factors (temperature, relative humidity) and COVID-19 case numbers across diverse climatic zones.

Main Methods:

• Employed three univariate Long Short-Term Memory (LSTM) variants (basic, stacked, bi-directional) for COVID-19 case prediction.
• Utilized a multivariate LSTM model incorporating temperature (T) and relative humidity (RH) as additional input variables.
• Analyzed data from 9 cities across India, USA, and Sweden during periods of minimal lockdown to observe uninhibited spread.

Main Results:

• The multivariate LSTM model, including environmental parameters, demonstrated superior performance compared to univariate models.
• An average improvement of 64% in Mean Absolute Percentage Error (MAPE) was observed with the inclusion of T and RH.
• Temperature showed varied correlations (positive in cold, negative in warm regions); RH correlations were mixed, influenced by local factors.

Conclusions:

• Integrating environmental parameters like temperature and relative humidity significantly enhances the accuracy of LSTM models for COVID-19 case forecasting.
• The findings suggest that environmental factors play a notable role in COVID-19 transmission dynamics.
• While beneficial, forecasting accuracy can be influenced by other confounding factors not included in the model.