Process management in diabetes treatment by blending technique.

Area of Science:

• Medical Informatics
• Artificial Intelligence in Healthcare
• Diabetes Management

Background:

• Diabetes mellitus is a chronic condition with significant long-term health risks affecting multiple organs.
• Effective diabetes management is crucial for preventing severe complications.
• Personalized health records offer valuable data for understanding and managing chronic diseases.

Purpose of the Study:

• To predict blood glucose levels in individuals with diabetes using advanced AI and Machine Learning (ML).
• To identify key factors influencing blood glucose levels and diabetes management.
• To evaluate the potential for AI-driven insights to improve patient outcomes and treatment personalization.

Main Methods:

• Utilized a large dataset (86,115 records) from E-Nabız personal health records, including lab results, medical history, and medication data.
• Applied feature selection techniques (SFM, MI, RFE, CHI2, ANOVA, KW, CATB, XGB, LGBM) to identify the top 20 relevant features.
• Employed a blending ensemble technique with CATB, XGB, and LGBM as base models and ETC as a meta-model for prediction and evaluation.

Main Results:

• The blending approach achieved high performance metrics: 92.52% precision, 92.51% recall, 92.51% F1-score, and 92.50% accuracy.
• Identified critical factors influencing diabetes management, including age, specific medications, and cholesterol levels.
• Demonstrated the superiority of the advanced ensemble method over single models or traditional ensemble techniques.

Conclusions:

• Advanced AI and ML, particularly ensemble methods, can effectively predict blood glucose levels and identify diabetes management indicators.
• The findings highlight the importance of integrating diverse data sources and sophisticated analytical techniques for personalized diabetes care.
• This research enhances the practical application of AI in clinical settings, supporting healthcare professionals in monitoring patients and tailoring treatments for better outcomes.