Mobile Application and Machine Learning-Driven Scheme for Intelligent Diabetes Progression Analysis and Management...

Area of Science:

• Digital health and machine learning applications in chronic disease management.
• Biomarker analysis for metabolic disorder monitoring.

Background:

• Diabetes mellitus affects over 500 million globally, requiring personalized management.
• Glycated hemoglobin (HbA1c) is a key biomarker for monitoring long-term glycemic control and diabetes progression.

Purpose of the Study:

• To develop an innovative approach for diabetes management by integrating a mobile application and machine learning.
• To analyze the dynamics of HbA1c levels and identify influencing factors for personalized diabetes care.

Main Methods:

• Creation of the DiabMini dataset with 127 features from 88 diabetic patients, including medical, personal, nutrient intake, and lifestyle data.
• Development of a stacking machine learning model (XGBoost, SVC, ET, KNN) to predict HbA1c dynamics.
• Application of SHapley Additive exPlanations (SHAP) for visualizing the impact of risk factors on HbA1c.

Main Results:

• The developed stacking model achieved a classification accuracy of 94.23% in predicting HbA1c dynamics.
• SHAP analysis effectively visualized and clarified the differential contributions of various risk factors to HbA1c levels.
• The study successfully created a novel, comprehensive diabetes dataset (DiabMini).

Conclusions:

• Integrating mobile health applications with machine learning offers a powerful tool for enhancing personalized diabetes management.
• The intelligent system provides insights into factors affecting HbA1c dynamics, facilitating tailored interventions.
• This approach holds significant potential for improving long-term diabetes control and patient outcomes.