Machine Learning-Based Prediction Model Construction for Type 2 Diabetes Mellitus: A Comparison of Algorithms and...

Area of Science:

• Public Health
• Biomedical Informatics
• Epidemiology

Background:

• Global high incidence of Type 2 diabetes mellitus (T2DM) necessitates improved prediction models.
• Existing models often lack multilevel integrated analysis of risk factors.
• Accurate T2DM prediction is crucial for individual health and healthcare systems.

Purpose of the Study:

• To construct a T2DM prediction model using machine learning.
• To identify multilevel risk factors for T2DM.
• To enable early screening and inform public health interventions.

Main Methods:

• Utilized National Health and Nutrition Examination Survey (NHANES) data (2021-2023).
• Employed advanced data preprocessing techniques including imputation, PCA, RF, RFE, and ADASYN.
• Compared performance of seven machine learning models, including AdaBoost.

Main Results:

• The AdaBoost model achieved optimal performance (AUC=0.85, Accuracy=0.71, F1=0.71).
• Identified 24 key risk factors across individual traits, behaviors, and living conditions.
• Model performance improved after parameter optimization.

Conclusions:

• Machine learning models integrating multidimensional risk factors offer accurate T2DM risk prediction.
• The health ecology framework combined with machine learning provides a scientific basis for multilevel interventions.
• This study offers a novel, comprehensive tool for precise T2DM prevention and public health strategies.