Comparative approaches for classification of diabetes mellitus data: Machine learning paradigm

Area of Science:

• Medical Informatics
• Machine Learning
• Statistical Modeling

Background:

• Diabetes affects millions globally, posing a significant financial and health burden.
• Accurate diabetes prediction is crucial for reducing costs and improving patient outcomes.
• Existing classification methods struggle with the non-normality and non-linearity of medical data.

Purpose of the Study:

• To evaluate Gaussian process classification (GPC) for diabetes prediction.
• To compare GPC performance against traditional classifiers like LDA, QDA, and NB.
• To analyze diabetes data using cross-validation and interpret the results.

Main Methods:

• Adapted Gaussian process (GP)-based classification with linear, polynomial, and radial basis kernels.
• Evaluated GP performance against Linear Discriminant Analysis (LDA), Quadratic Discriminant Analysis (QDA), and Naive Bayes (NB).
• Utilized metrics including accuracy (ACC), sensitivity (SE), specificity (SP), positive predictive value (PPV), and negative predictive value (NPV).

Main Results:

• The GP-based model achieved an accuracy of 81.97% on the Pima Indian diabetes dataset.
• GP demonstrated superior sensitivity (91.79%) and PPV (84.91%) compared to other methods.
• The study analyzed 768 patients, with 268 diagnosed as diabetic.

Conclusions:

• Gaussian process classification offers a robust and effective method for diabetes prediction.
• GP-based models show potential for improving diagnostic accuracy in complex medical datasets.
• Further research can explore GP's application in other challenging classification tasks within healthcare.