A Fusion-Based Machine Learning Approach for the Prediction of the Onset of Diabetes

Area of Science:

• Healthcare AI
• Machine Learning in Medicine
• Computational Biology

Background:

• Machine learning applications in healthcare face challenges due to limited, low-quality data, impacting diagnostic model accuracy.
• Diabetes is a prevalent chronic condition with significant global healthcare costs.
• Accurate identification and prediction of critical events are crucial for managing patients with diabetes (PwD).

Purpose of the Study:

• To present a fusion machine learning approach for improved diabetes identification and prediction of critical events in patients with diabetes (PwD).
• To enhance the accuracy of machine learning models in the healthcare domain, specifically for diabetes management.
• To provide a foundation for personalized treatment strategies, potentially increasing survival rates for PwD.

Main Methods:

• Development of a fusion machine learning architecture.
• Integration of Support Vector Machine and Artificial Neural Network classifiers.
• Training and validation using healthcare data for diabetes-related predictions.

Main Results:

• Achieved a classification accuracy of 94.67% for diabetes identification and event prediction.
• Demonstrated a ~1.8% improvement over the best previously reported machine learning models for diabetes.
• The fusion approach showed superior performance compared to individual models.

Conclusions:

• The proposed fusion machine learning approach significantly improves the accuracy of diabetes identification and critical event prediction.
• This enhanced accuracy has the potential to inform optimal, individualized treatment plans for patients with diabetes (PwD).
• The solution offers a promising tool to support increased survival rates and reduce the burden of diabetes on healthcare systems.