Diabetes mellitus prediction and diagnosis from a data preprocessing and machine learning perspective.

Area of Science:

• Medical Informatics
• Computational Biology
• Machine Learning in Healthcare

Background:

• Diabetes mellitus is a critical health issue characterized by hyperglycemia, posing life-threatening risks if unmanaged.
• Machine learning offers a promising avenue for diabetes prediction, but current accuracy rates necessitate further improvement.
• Effective computational diagnosis of diabetes is crucial for timely intervention and patient management.

Purpose of the Study:

• To propose a novel machine learning framework for enhanced diabetes prediction and diagnosis.
• To investigate the impact of feature selection and missing value imputation on model performance.
• To develop a robust model for accurate classification of diabetes mellitus.

Main Methods:

• Implemented Spearman correlation for feature selection and polynomial regression for missing value imputation.
• Employed supervised machine learning models: Random Forest (RF), Support Vector Machine (SVM), and a novel twice-growth deep neural network (2GDNN).
• Utilized grid search and repeated stratified k-fold cross-validation for hyperparameter optimization.

Main Results:

• The proposed 2GDNN model achieved high performance metrics, including 97.34% precision and 97.25% test accuracy on the PIMA Indian dataset.
• On the LMCH diabetes dataset, the 2GDNN model attained 97.28% precision and 97.33% test accuracy.
• The framework demonstrated superior performance compared to existing state-of-the-art methods in diabetes prediction.

Conclusions:

• The developed machine learning framework, incorporating advanced preprocessing and optimized classifiers, yields a robust model for diabetes mellitus prediction.
• The proposed methods significantly outperform current state-of-the-art results in diabetes diagnosis.
• Source code for the developed models is publicly available to facilitate further research and application.