Diabetes prediction model based on GA-XGBoost and stacking ensemble algorithm.

Area of Science:

• Medical Informatics
• Computational Biology
• Public Health

Background:

• Diabetes is a chronic, incurable disease requiring early intervention for better patient outcomes.
• Accurate early diagnosis and personalized treatment are crucial for managing diabetes effectively.

Purpose of the Study:

• To develop an advanced machine learning model for early diabetes risk prediction.
• To provide a scientific basis for timely diagnosis and treatment of diabetes.

Main Methods:

• Utilized the Behavioral Risk Factor Surveillance System (BRFSS) dataset.
• Applied data balancing techniques, including SMOTEENN, to address data imbalance.
• Constructed a Stacking model integrating a genetic algorithm-optimized XGBoost (GA-XGBoost) with LightGBM and random forest models.
• Employed Shapley values for model interpretability and feature importance analysis.

Main Results:

• SMOTEENN demonstrated superior performance in data balancing.
• The GA-XGBoost model enhanced predictive accuracy through hyperparameter optimization.
• The two-layer Stacking model outperformed individual machine learning models in predictive efficacy.
• Shapley value analysis identified age and body mass index as significant predictors of diabetes risk.

Conclusions:

• The integrated Stacking model offers a novel and effective approach for diabetes risk prediction.
• Model interpretability through Shapley values aids in clinical decision-making and personalized treatment.
• This study provides a powerful tool for early diabetes diagnosis and tailored patient care.