COVID-19 mortality and nutrition through predictive modeling and optimization based on grid search

Area of Science:

• Epidemiology
• Nutritional Science
• Computational Biology

Background:

• COVID-19, particularly the Omicron variant, has caused millions of deaths globally.
• Nutrition plays a critical role in disease resistance and health outcomes.
• Dietary components like fat, protein, and vegetables are linked to COVID-19 severity and mortality.

Purpose of the Study:

• To investigate the relationship between dietary intake and COVID-19 death rates.
• To evaluate the effectiveness of machine learning models in predicting nutrition-related disease resistance.
• To identify the optimal machine learning model for predicting COVID-19 mortality based on nutritional data.

Main Methods:

• Utilized a COVID-19 nutrition dataset with attributes including fat percentage, caloric intake, food supply amount, and protein levels.
• Applied five machine learning models: Gradient Boosting Regressor (GBR), Random Forest (RF), Lasso Regression, Decision Tree (DT), and Bayesian Ridge (BR).
• Employed Grid Search (GS) for hyperparameter optimization of the GBR model and evaluated models using R², MAE, MAPE, and MSE.

Main Results:

• The Gradient Boosting Regressor (GBR) model initially showed optimal performance among the tested models.
• The unoptimized GBR achieved a R² of 0.963, MSE of 0.1512, MAE of 0.2262, and MAPE of 0.1351.
• Grid Search optimization significantly enhanced the GBR model's performance, increasing its R² to 0.994.

Conclusions:

• The Grid Search-optimized Gradient Boosting Regressor (GS-GBR) demonstrates superior predictive accuracy for COVID-19 mortality rates.
• Nutritional factors are significant predictors of COVID-19 outcomes.
• Machine learning models, particularly optimized GBR, can improve predictions for nutrition-related disease resistance.