A data-driven interpretable ensemble framework based on tree models for forecasting the occurrence of COVID-19 in the...

Area of Science:

• Epidemiology
• Computational Biology
• Public Health

Background:

• The COVID-19 pandemic presents a significant global public health challenge.
• Tree-based machine learning offers efficient and interpretable disease prediction capabilities.

Purpose of the Study:

• To develop an interpretable ensemble framework using tree-based models to forecast daily COVID-19 new cases in the USA.
• To identify key factors influencing COVID-19 transmission and control.

Main Methods:

• Developed and optimized three machine learning algorithms: Random Forest (RF), eXtreme Gradient Boosting (XGBoost), and Light Gradient Boosting Machine (LightGBM).
• Integrated model outcomes using three linear ensemble models for enhanced prediction accuracy.
• Utilized SHapley Additive explanation (SHAP) values for feature importance ranking.

Main Results:

• LightGBM demonstrated the highest prediction accuracy among the base learners, followed by XGBoost and RF.
• The optimized LAD ensemble model achieved the highest precision, reducing prediction error by approximately 3.111% compared to the best base learner (LightGBM).
• Identified vaccination, mask-wearing, reduced mobility, and government interventions as significant factors positively impacting COVID-19 control.

Conclusions:

• An interpretable ensemble framework effectively forecasts COVID-19 cases.
• Machine learning models, particularly LightGBM and ensemble approaches, show strong predictive power for infectious diseases.
• Public health interventions like vaccination and mask mandates are crucial for pandemic management.