Predictive modeling of multi-class diabetes mellitus using machine learning and filtering iraqi diabetes data...

Area of Science:

• Medical Informatics
• Machine Learning
• Data Science

Background:

• Diabetes mellitus is a chronic metabolic disorder characterized by hyperglycemia.
• Poorly managed diabetes leads to severe complications affecting nerves, blood vessels, and organs.
• Accurate detection and classification of diabetes are crucial for timely intervention and management.

Purpose of the Study:

• To propose a multiclass diabetes mellitus detection and classification approach.
• To address challenges posed by extremely imbalanced datasets from the Laboratory of Medical City Hospital.
• To develop and evaluate a new, moderately imbalanced dataset derived from the same source.

Main Methods:

• Utilized three machine learning classifiers: Support Vector Machine (SVM), Logistic Regression, and K-Nearest Neighbor (KNN).
• Employed dimensionality reduction techniques including filter, wrapper, and embedded feature selection methods.
• Optimized classifier performance through hyperparameter tuning using 10-fold grid search cross-validation.

Main Results:

• Achieved a maximum accuracy of 0.964 and AUC of 0.99 with SVM on the original imbalanced dataset using top 4 features (filter method).
• KNN classifier achieved 0.935 accuracy and 1.0 for other metrics using top 9 features (wrapper method).
• SVM classifier attained 0.938 accuracy and 1.0 for other metrics on the newly created moderately imbalanced dataset.

Conclusions:

• The proposed machine learning approach effectively detects and classifies multiclass diabetes mellitus.
• The study demonstrates superior performance compared to existing research on the Laboratory of Medical City Hospital dataset.
• Feature selection and hyperparameter optimization are key to enhancing classification accuracy in imbalanced medical datasets.