Efficient diagnosis of diabetes mellitus using an improved ensemble method.

Area of Science:

• Computational biology and bioinformatics
• Medical informatics and health data science

Background:

• Diabetes mellitus presents a significant global health challenge, particularly in developing nations, with high mortality rates.
• Existing machine learning (ML) models for diabetes detection often suffer from low accuracy due to overfitting, underfitting, and data noise.

Purpose of the Study:

• To improve the classification accuracy of diabetes detection using advanced machine learning techniques.
• To develop a robust predictive model for early diabetes mellitus identification.

Main Methods:

• Utilized the Pima India Diabetes Data from the UCI ML Repository.
• Applied data preprocessing including imputation of missing values and feature selection via forward and backward methods.
• Employed parallel and sequential ensemble machine learning approaches (Random Forest, XG Boost, AdaBoostM1, Gradient Boosting) with feature selection for binary classification.

Main Results:

• Achieved 100% classification accuracy with XG Boost, AdaBoostM1, and Gradient Boosting ensemble methods.
• All performance metrics, including F1 score, MCC, Precision, Recall, AUC-ROC, and AUC-PR, reached 1.00, indicating highly reliable predictions.
• The developed ensemble models demonstrated superior performance over traditional methods, overcoming common ML challenges.

Conclusions:

• Ensemble machine learning methods, particularly XG Boost, AdaBoostM1, and Gradient Boosting, combined with feature selection, significantly enhance diabetes prediction accuracy.
• The validated predictive model provides a reliable tool for rapid diabetes mellitus detection, with potential to aid clinical decision-making and improve patient outcomes.
• This research offers a scalable and effective approach for early diabetes diagnosis in resource-limited settings.