Impact of Feature Selection Techniques for the Prediction of Cardiovascular Disease

Heart disease is the primary cause of fatality worldwide, and the comprehensive clinical data fetched by medical devices play an important role in its prediction and management. However, the size and complexity of the data involved can lead to incorrect predictions at times. This work emphasizes the importance of feature selection (FS) for improving disease prediction accuracy in machine learning. The FS helps to identify and retain relevant parameters while removing unnecessary features. Therefore, seven FS techniques, namely chi-squared, mutual information (MI), relief, forward FS, recursive feature elimination (RFE), lasso, and ridge regression, are utilized for selecting the most relevant features from the Echocardiogram data. Furthermore, a widely accepted classifier, random forests (RF), is applied to predict cardiovascular disease (CVD). Results reveal that the feature subset selected by the Relief technique achieves the highest classification accuracy of 97.23% and precision of 97.19%, compared to other feature selection techniques.