Decision tree with randomized grid search-based hyperparameter tuning and optimal feature scaling for diabetes...

Area of Science:

• Medical Informatics
• Computational Biology
• Machine Learning

Background:

• Diabetes mellitus is a chronic metabolic disorder characterized by hyperglycemia.
• Uncontrolled diabetes leads to severe long-term complications affecting multiple organ systems.
• Machine learning (ML) is increasingly utilized for predicting diabetes onset from patient data.

Purpose of the Study:

• To develop and evaluate a robust, data-driven machine learning framework for enhanced diabetes classification.
• To improve the accuracy and efficiency of diabetes diagnosis using advanced ML techniques.
• To provide a scalable, interpretable, and ethically aligned tool for early diabetes detection.

Main Methods:

• A two-stage decision tree (DT) model employing randomized grid search for hyperparameter tuning.
• Evaluation of eight feature scaling methods (e.g., StandardScaler, MinMaxScaler) to identify optimal data preprocessing.
• Utilized 5-fold cross-validation and metrics like accuracy, F1-score, AUC, and MCC on PIMA and IPDD datasets.

Main Results:

• Achieved high classification performance, with metrics reaching up to 99.8% on the IPDD dataset.
• Demonstrated strong performance on the PIMA dataset, with metrics around 77%.
• Compared favorably against various ML classifiers, including classic DT, KNN, SVM, and ensemble methods.

Conclusions:

• The proposed ML framework offers a significant advancement in diabetes diagnosis.
• The tool is scalable, interpretable, and ethically aligned, aiding early disease detection.
• Contributes to reducing the health risks associated with undiagnosed or poorly managed diabetes.