Exploring feature selection and classification methods for predicting heart disease.

Area of Science:

• Cardiology
• Machine Learning
• Data Science

Background:

• Machine learning (ML) significantly improves computer-aided diagnosis (CAD) accuracy.
• Effective feature selection is crucial for optimizing ML model performance in healthcare.

Purpose of the Study:

• To experimentally assess ML model performance using various feature selection methods for heart disease datasets.
• To identify optimal feature combinations for enhanced heart condition prediction accuracy.

Main Methods:

• Evaluated four heart disease datasets using Principal Component Analysis (PCA), Chi-squared testing, ReliefF, and Symmetrical Uncertainty for feature selection.
• Developed and compared various classification models using distinct feature sets to determine optimal combinations.

Main Results:

• The effectiveness of feature selection varied across different ML techniques and heart datasets.
• The optimal model, utilizing Chi-squared feature selection with the BayesNet algorithm, achieved a prediction accuracy of 85.00%.

Conclusions:

• Feature selection strategies can significantly impact the accuracy of ML-based heart disease diagnostic systems.
• The combination of Chi-squared feature selection and BayesNet offers a promising approach for accurate heart condition prediction.