Robust predictive framework for diabetes classification using optimized machine learning on imbalanced datasets

Area of Science:

• Medical Informatics
• Machine Learning
• Data Science

Background:

• Clinical data analysis for diabetes prediction is vital but challenged by class imbalance.
• Dominance of non-diabetic cases in datasets leads to biased machine learning models and poor generalization.
• Accurate diabetes prediction is essential for timely medical intervention and patient management.

Purpose of the Study:

• To develop and evaluate a novel predictive framework for diabetes prediction.
• To address the critical issue of class imbalance in clinical datasets.
• To enhance the accuracy and generalizability of machine learning models for diabetes diagnosis.

Main Methods:

• Developed a novel predictive framework integrating advanced machine learning algorithms.
• Employed cutting-edge imbalance handling techniques, including feature engineering and resampling strategies.
• Tested the framework's robustness and adaptability on three diverse datasets: PIMA, Diabetes Dataset 2019, and BIT_2019.

Main Results:

• The framework demonstrated robust performance across different datasets, effectively handling class imbalance.
• Model selection and imbalance mitigation strategies were shown to be critical for reliable predictions.
• The methodology proved adaptable to varying data environments, confirming its practical utility.

Conclusions:

• The proposed data-driven framework significantly advances diabetes prediction accuracy by effectively tackling class imbalance.
• This study underscores the importance of specialized techniques for improving machine learning model performance in medical informatics.
• The findings offer a valuable contribution to the field, paving the way for more reliable and generalizable diagnostic tools.