Optimal machine learning models and factors for global infectious diseases: An ecological analysis

Infectious diseases are a major global public health challenge. Socioeconomic and environmental changes have complicated the situation, underscoring the need for effective predictive models and key factor identification for prevention and control. The study aims to explore the establishment of an optimal machine learning model adaptation system for infectious diseases to provide standardized data support and reusable model references for related follow-up research.

This study included infectious disease incidence data from the Global Burden of Disease 2021 database, as well as 19 factors including socioeconomic and air pollution indicators from the Our World in Data platform. Eight machine learning models were included, such as Gradient Boosting Decision Tree (GBDT), Light Gradient Boosting Machine (LightGBM), and Random Forest (RF). Model performance was evaluated using five metrics, including the Coefficient of Determination (R²). The SHapley Additive exPlanations (SHAP) method was applied to quantify the contributions of factors at both global and regional levels.

The GBDT performed best for sexually transmitted (R² = 0.506) and respiratory infectious diseases (R² = 0.854); RF was optimal for all infectious diseases (R² = 0.662); LightGBM suited enteric infectious diseases (R² = 0.725); GBDT also achieved the optimal performance for neglected tropical diseases (R² = 0.434). SHAP identified core factors: child dependency ratio (SHAP = 0.47 for diarrheal diseases), unsafe drinking water (SHAP = 0.23 for all infectious diseases), particulate matter 2.5 exposure (SHAP = 0.19 for respiratory infectious diseases), and socio-demographic index (SHAP = 0.17 for other types).

Adapted machine learning models efficiently identify key factors. Core strategies to reduce the global infectious disease burden include improving drinking water safety, promoting clean energy, optimizing diets, and strengthening child health services.