Neural Network-Based Coronary Heart Disease Risk Prediction Using Feature Correlation Analysis

Area of Science:

• Cardiology
• Machine Learning
• Biostatistics

Background:

• Neural networks (NNs) are widely used for coronary heart disease (CHD) risk prediction.
• However, the "black-box" nature of NNs limits medical expert satisfaction with their predictive performance.

Purpose of the Study:

• To develop an improved NN-based prediction model for CHD risk.
• To enhance model interpretability and predictive accuracy through feature correlation analysis.

Main Methods:

• A two-stage approach was employed: feature selection based on predictive importance and feature correlation analysis.
• The proposed model, Neural Network-Feature Correlation Analysis (NN-FCA), was developed and validated.

Main Results:

• The NN-FCA model achieved a significantly higher area under the receiver operating characteristic (ROC) curve (0.749 ± 0.010) compared to the Framingham risk score (FRS) (0.393 ± 0.010).
• The study evaluated 4146 individuals, with 3031 classified as low CHD risk and 1115 as high risk.

Conclusions:

• The NN-FCA model offers superior performance in predicting CHD risk compared to the FRS.
• The proposed method provides more accurate CHD risk predictions and improved ROC curve performance in the Korean population.