Robust diabetic prediction using ensemble machine learning models with synthetic minority over-sampling technique

Area of Science:

• Medical Informatics
• Machine Learning
• Public Health

Background:

• Diabetes mellitus is a global health crisis characterized by insulin resistance or deficiency, leading to hyperglycemia.
• Elevated blood sugar levels can cause severe complications, including kidney disease, vision loss, and cardiovascular conditions.
• Early diagnosis is critical for managing diabetes and preventing its severe health consequences.

Purpose of the Study:

• To develop and validate a robust machine learning framework for accurate diabetes prediction.
• To enhance the performance of predictive models through data preprocessing and ensemble techniques.
• To improve early detection rates for better diabetes management and patient outcomes.

Main Methods:

• Data preprocessing included imputation of missing values and outlier rejection.
• Feature selection was performed using correlation analysis.
• Class distribution was balanced using the Synthetic Minority Over-sampling Technique (SMOTE).
• Ensemble machine learning models, specifically AdaBoost and XGBoost, were employed for prediction.

Main Results:

• The proposed AdaBoost and XGBoost ensemble model achieved an Area Under the Curve (AUC) of 0.968 +/- 0.015.
• This performance surpassed alternative methods evaluated in the study.
• The model demonstrated state-of-the-art results in diabetes prediction accuracy.

Conclusions:

• The developed framework offers a highly effective approach for early diabetes prediction.
• The integration of SMOTE with ensemble methods significantly boosts predictive performance.
• This model holds promise for advancing diabetes management and improving healthcare outcomes.