Two Machine-learning Hybrid Models for Predicting Type 2 Diabetes Mellitus

Area of Science:

• Medical Informatics
• Computational Biology
• Machine Learning in Healthcare

Background:

• Rising global diabetes prevalence necessitates advanced diagnostic tools.
• Machine learning (ML) shows significant potential for disease diagnosis, including diabetes.
• Accurate diabetes diagnosis is crucial for timely intervention and management.

Purpose of the Study:

• To develop and evaluate novel hybrid machine learning models for diabetes diagnosis.
• To address the challenge of outlier data in medical datasets for improved diagnostic accuracy.
• To compare the performance of proposed hybrid models against established ML algorithms.

Main Methods:

• Utilized a dataset of 1000 physical examination samples from Iraqi hospitals, categorized as diabetic, nondiabetic, and predicted diabetic.
• Developed Hybrid Model 1 (K-medoids + Gaussian Naive Bayes with KDE) to handle outliers.
• Developed Hybrid Model 2 (K-means + Gaussian Naive Bayes with KDE) for data with outliers removed.
• Applied Principal Component Analysis for dimensionality reduction and fivefold cross-validation for evaluation.

Main Results:

• Proposed hybrid models demonstrated superior performance in diabetes prediction compared to Support Vector Machines, Decision Trees (J48), and Gaussian Naive Bayes.
• Hybrid Model 1 achieved an average accuracy of 0.9743 in handling outlier data.
• Hybrid Model 2 achieved an average accuracy of 0.9867 after outlier removal.
• Precision, sensitivity, and F1-score were also evaluated, indicating robust performance.

Conclusions:

• The study successfully presented two accurate hybrid models for diabetes diagnosis.
• The models effectively managed outlier data, enhancing diagnostic reliability.
• Findings underscore the potential of ML for improving early diabetes diagnosis and treatment strategies.