Related Experiment Video
Updated: May 11, 2026
12:18
A Machine Learning Approach to Design an Efficient Selective Screening of Mild Cognitive Impairment
Published on: January 11, 2020
7.5K
Balancing Acts: Tackling Data Imbalance in Machine Learning for Predicting Myocardial Infarction in Type 2 Diabetes
Berk Ozturk1, Tom Lawton1,2, Stephen Smith1
1University of York, York, YO10 5GH, UK.
Studies in Health Technology and Informatics
|August 23, 2024
Summary
Class imbalance in machine learning models for predicting Type 2 Diabetes (T2D) complications like heart attack (Myocardial Infarction) can be overcome. Class Weighting (CW) significantly improved model performance, offering a promising approach for early risk detection in T2D patients.
Area of Science:
- Medical Informatics
- Cardiology
- Machine Learning
Background:
- Type 2 Diabetes (T2D) affects over 500 million adults globally, increasing the risk of severe comorbidities.
- Myocardial Infarction (MI), or heart attack, is a critical and life-threatening comorbidity associated with T2D.
- Accurate prediction of MI in T2D patients is crucial for timely intervention and improved patient outcomes.
Purpose of the Study:
- To investigate the impact of Class Imbalance Handling (CIH) techniques on Machine Learning (ML) model performance for predicting MI in T2D patients.
- To evaluate the effectiveness of Oversampling, Undersampling, and Class Weighting (CW) in addressing class imbalance challenges.
- To identify the most effective CIH technique for enhancing ML model accuracy in this clinical prediction task.
Main Methods:
- Utilized the Connected Bradford dataset, containing over one million real-world health records.
- Applied three common CIH techniques: Oversampling, Undersampling, and Class Weighting (CW).
- Evaluated CIH techniques across various ML models including Naive Bayes (NB), Neural Network (NN), Random Forest (RF), Support Vector Machine (SVM), and Ensemble models.
Main Results:
- Class Weighting (CW) demonstrated superior performance compared to other CIH techniques.
- CW achieved the highest Accuracy (0.9948) and F1 score (0.9556) among the evaluated methods.
- Systematic application of CIH techniques significantly improved ML model performance for MI prediction in T2D.
Conclusions:
- Effective CIH techniques, particularly CW, can substantially enhance the reliability and accuracy of ML models for clinical predictions.
- Implementing appropriate CIH strategies in ML models using real-world healthcare data shows significant promise for reducing MI risk in T2D patients.
- This research highlights the importance of addressing class imbalance for trustworthy AI deployment in healthcare.
