Predicting diabetes mellitus using SMOTE and ensemble machine learning approach: The Henry Ford ExercIse Testing...

Area of Science:

• Medical research
• Machine learning applications in healthcare
• Cardiorespiratory fitness and metabolic disease prediction

Background:

• Machine learning is increasingly vital in medical research.
• Predicting incident diabetes is crucial for public health.
• Cardiorespiratory fitness data offers a potential source for diabetes prediction.

Purpose of the Study:

• To evaluate machine learning methods for predicting incident diabetes.
• To identify potential predictors of diabetes using patient data.
• To assess the performance of ensemble models in diabetes prediction.

Main Methods:

• Utilized data from 32,555 patients undergoing exercise treadmill stress testing.
• Applied various machine learning algorithms including Decision Tree, Naïve Bayes, Logistic Regression, Logistic Model Tree, and Random Forests.
• Developed an ensemble model using 13 selected attributes, addressing class imbalance with Synthetic Minority Oversampling Technique (SMOTE).

Main Results:

• An ensemble model combining Naïve Bayes Tree, Random Forest, and Logistic Model Tree achieved high prediction accuracy (AUC = 0.92).
• Ensemble methods and SMOTE demonstrated effectiveness in handling class imbalance and improving predictive performance.
• Identified key attributes contributing to diabetes prediction through clinical importance, Multiple Linear Regression, and Information Gain Ranking.

Conclusions:

• Ensemble machine learning approaches show significant potential for predicting incident diabetes.
• Cardiorespiratory fitness data, combined with machine learning, can aid in early diabetes detection.
• SMOTE is an effective technique for managing class imbalance in predictive models for diseases like diabetes.