Machine learning models for classification and identification of significant attributes to detect type 2 diabetes

Area of Science:

• Computational biology
• Medical informatics

Background:

• Type 2 Diabetes (T2D) is a complex chronic condition marked by insulin resistance and impaired insulin secretion.
• Accurate subtyping of T2D is crucial for effective prognosis and tailored treatment strategies.

Purpose of the Study:

• To apply machine learning (ML) techniques for categorizing T2D patients.
• To identify key T2D-associated features for improved prediction models.

Main Methods:

• Utilized the Pima Indian Diabetes Dataset for analysis.
• Employed feature selection and various classification techniques.
• Evaluated classifier performance using accuracy, kappa statistics, AUROC, sensitivity, specificity, and logloss.

Main Results:

• Generalized Boosted Regression achieved 90.91% accuracy.
• Several models demonstrated 100% sensitivity and high AUROC (95.26%).
• Glucose levels, BMI, diabetes pedigree function, and age were identified as significant predictors.

Conclusions:

• ML methods are effective in developing robust prediction models for T2D.
• Identified key clinical and demographic features for T2D outcome prediction in the Pima Indian population.