Diabetes Prediction Using Feature Selection Algorithms and Boosting-Based Machine Learning Classifiers.

Area of Science:

• Medical Informatics
• Computational Biology
• Machine Learning in Healthcare

Background:

• Diabetes mellitus is a major global health issue requiring early diagnosis to prevent complications.
• Accurate diabetes prediction is hindered by limited, noisy, and imbalanced datasets.
• Existing methods face challenges in feature selection, class imbalance, and data preprocessing.

Purpose of the Study:

• To develop a novel machine learning framework for enhanced diabetes prediction.
• To address challenges of feature selection, class imbalance, and data preprocessing.
• To improve the accuracy and interpretability of early diabetes detection models.

Main Methods:

• Systematic evaluation of five feature selection algorithms: Recursive Feature Elimination, Grey Wolf Optimizer, Particle Swarm Optimizer, Genetic Algorithm, and Boruta.
• Utilized cross-validation and SHAP analysis for feature selection and interpretability.
• Employed LightGBM (LGBM) and XGBoost (XGBoost) classification algorithms.

Main Results:

• The Boruta feature selection algorithm identified the top five features, leading to superior performance with the LightGBM classifier.
• Achieved 85.16% accuracy and an 85.41% F1-score.
• Demonstrated a 54.96% reduction in training time compared to other configurations.

Conclusions:

• The proposed framework provides a robust and accurate solution for early diabetes detection.
• Validated effectiveness on the Pima Indian Diabetes Dataset and the DiaHealth dataset.
• Offers a cost-effective, interpretable, and clinically relevant approach for early diabetes detection.