Related Experiment Videos
Optimizing stability of heart disease prediction across imbalanced learning with interpretable Grow Network
Simon Bin Akter1, Sumya Akter1, Rakibul Hasan2
1Martin Tuchman School of Management, New Jersey Institute of Technology, Newark, 07102, NJ, USA; Department of Computer Science and Engineering, Northern University Bangladesh, Dhaka, Bangladesh.
Computer Methods and Programs in Biomedicine
|March 27, 2025
Summary
This study introduces GrowNet, a stable heart disease prediction model that excels with imbalanced public datasets. GrowNet improves early detection by identifying key risk factors, enhancing patient outcomes.
Area of Science:
- Cardiology
- Machine Learning
- Data Science
Background:
- Heart disease prediction models struggle with stability on imbalanced public datasets.
- Class imbalances negatively impact feature selection, sampling, and modeling, leading to biased performance.
Purpose of the Study:
- To develop a stable and interpretable heart disease prediction model.
- To address the challenges posed by class imbalance in real-world datasets.
Main Methods:
- Proposed a novel Grow Network (GrowNet) architecture with dynamic data configuration.
- Implemented TriDyn Dependence feature selection and Adaptive Refinement sampling for stability.
- Utilized SHapley Additive exPlanations (SHAP) for model interpretability.
Main Results:
- GrowNet outperformed other models on the UCI heart disease dataset (Specificity: 92%, Sensitivity: 88%).
- Demonstrated robust performance on imbalanced BRFSS and NHIS datasets, outperforming other models.
- Achieved significant improvements in handling class imbalance, showing enhanced stability and generalizability.
Conclusions:
- GrowNet offers a stable and interpretable solution for early heart disease detection.
- The model assists healthcare professionals in identifying key risk factors, improving patient outcomes.
- Facilitates cost-effective heart disease prediction through focus on critical risk indicators.
