Machine Learning Models for Data-Driven Prediction of Diabetes by Lifestyle Type

Area of Science:

• Computational biology
• Public health
• Data science

Background:

• Diabetes prevalence is rising globally.
• Machine learning (ML) models show promise for diabetes prediction.
• Lifestyle factors significantly influence diabetes risk.

Purpose of the Study:

• To compare the predictive efficacy of five ML models for diabetes.
• To identify key lifestyle variables for diabetes prediction.
• To leverage the National Health and Nutrition Examination Survey (NHANES) database.

Main Methods:

• Utilized the 1999-2020 NHANES database (17,833 participants).
• Employed the Akaike Information Criterion (AIC) forward algorithm for data screening.
• Developed and evaluated five ML models: CATBoost, XGBoost, Random Forest (RF), Logistic Regression (LR), and Support Vector Machine (SVM).
• Assessed model performance using accuracy, sensitivity, specificity, precision, F1 score, and ROC curves.

Main Results:

• Dietary intake (energy, carbohydrate, fat) were the most significant predictors.
• CATBoost outperformed RF, LR, XGBoost, and SVM.
• CATBoost achieved the highest accuracy (82.1%) and AUC (0.83).

Conclusions:

• ML models, particularly CATBoost, can accurately predict diabetes risk.
• NHANES data and ML offer a valuable tool for early diabetes detection in clinical settings.
• Dietary factors are crucial for diabetes prediction models.