Predicting COVID-19 cases in various scenarios using RNN-LSTM models aided by adaptive linear regression to identify...

Area of Science:

• Epidemiology and Public Health
• Data Science and Artificial Intelligence
• Socioeconomic Determinants of Health

Background:

• The COVID-19 pandemic highlighted complex issues in social inequality, poverty, and healthcare systems.
• Understanding virus evolution and infection spread is crucial for effective public health interventions.
• Previous analyses often overlooked data anomalies and the specific challenges in under-resourced regions.

Purpose of the Study:

• To evaluate the predictive accuracy of Recurrent Neural Network-Long Short Term Memory (RNN-LSTM) for daily COVID-19 infection fluctuations.
• To identify and correct data anomalies using adaptive linear regression for a more reliable epidemiological scenario.
• To support evidence-based epidemic control actions, particularly in areas with limited testing and prevention.

Main Methods:

• Utilized daily COVID-19 infection data for predictive modeling.
• Implemented Recurrent Neural Network-Long Short Term Memory (RNN-LSTM) for time-series forecasting.
• Applied adaptive linear regression to detect and adjust for anomalies in reported case data.

Main Results:

• The RNN-LSTM model demonstrated significant predictive power for daily COVID-19 infection dynamics.
• Adaptive linear regression successfully identified and corrected anomalies, offering a more realistic view of the pandemic's spread.
• The methodology proved particularly effective in scenarios with low testing, diagnosis, and prevention rates.

Conclusions:

• The combined approach of RNN-LSTM and adaptive linear regression enhances the reliability of COVID-19 data analysis.
• This method is valuable for epidemiological surveillance and control, especially in vulnerable populations and regions like Brazil's favelas.
• Accurate data analysis is essential for targeted interventions to mitigate infection and death rates during pandemics.