An intelligent diabetes classification and perception framework based on ensemble and deep learning method.

Area of Science:

• Biomedical Informatics
• Computational Biology
• Machine Learning in Healthcare

Background:

• Elevated blood sugar levels pose significant health risks, including blindness, renal, kidney, and heart diseases.
• Diabetes mellitus contributes to a substantial global mortality rate, with diabetic patients facing an average annual mortality of 38%.

Purpose of the Study:

• To develop and evaluate machine learning classifiers for early diabetes onset prediction.
• To identify key features for accurate diabetes risk assessment.

Main Methods:

• Employed Chi-square, mutual information, and sequential feature selection (SFS) for feature selection.
• Trained and compared multiple classifiers: artificial neural network (ANN), random forest (RF), gradient boosting (GB), Tab-Net, and support vector machine (SVM).
• Validated the system on the PIMA and early-risk diabetes datasets.

Main Results:

• The ANN achieved 99.35% accuracy on the PIMA dataset.
• The RF model attained 99.36% accuracy on the early diabetes risk dataset.
• The proposed method demonstrated superior performance compared to baseline machine learning algorithms for diabetes prediction.

Conclusions:

• The developed feature selection and classification approach enables accurate early diagnosis of diabetes.
• Machine learning models, particularly ANN and RF, show high efficacy in predicting diabetes risk.
• The findings suggest a significant advancement in automated diabetes prediction systems.