A novel framework for diabetic risk prediction using SCAW-Net integrated with TabNet architecture

Area of Science:

• Biomedical Informatics
• Machine Learning in Healthcare
• Endocrinology

Background:

• Blood glucose regulation by insulin is vital for metabolism and brain function.
• Effective diabetes management requires precise and rapid identification to prevent complications.
• Existing prediction models may lack accuracy or speed for clinical application.

Purpose of the Study:

• To introduce SCAW-Net integrated within the TabNet architecture for enhanced diabetes prediction.
• To evaluate the performance of SCAW-Net against traditional machine learning algorithms.
• To assess the model's efficacy on diverse and imbalanced datasets.

Main Methods:

• Development of SCAW-Net, a novel deep learning model, integrated into the TabNet framework.
• Training and testing the model using comprehensive diabetes-related features across multiple datasets.
• Comparative analysis against AdaBoost, XGBoost, Bagging, and Random Forest algorithms.

Main Results:

• SCAW-Net achieved a high prediction accuracy of 98.9%, surpassing other evaluated methods.
• The model demonstrated consistent performance on complex and imbalanced datasets.
• Significant improvements in prediction accuracy and computational speed were observed.

Conclusions:

• SCAW-Net within TabNet is a highly accurate and efficient tool for diabetes prediction.
• The model's robustness on challenging datasets supports its potential for real-world clinical use.
• This approach can facilitate timely interventions and improve diabetes patient management.