The Effect of Feature Subset Optimization on the Precision of Cardiovascular Disease Predictive Models

Abstract

In the last 10 years, heart disease has arisen as a significant worldwide health problem that substantially influences mortality rates. To avoid patients from any harm, prompt and accurate testing is vital. In order to recognize cardiovascular illness, non-invasive medical technologies—especially those that make use of artificial intelligence and machine learning—have become more crucial. However, heart illness prediction is still hard, particularly in light of the intricacy of vast volumes of medical data. With the purpose of boosting the accuracy of heart disease diagnosis, this study tries to find relevant risk indicators in high-dimensional datasets. Two distinct heart disease files with varied medical features were employed in order to accomplish this. The link and correlation between these features and heart disease was the principal focus of the investigation. After that, a filter-based feature selection strategy was incorporated to these datasets, providing a reduced feature group that may be utilized to identify heart disease. Many machine learning classification models that incorporated both reduced and complete feature groups were constructed for trial analysis. Learned models were assessed using precision, the Receiver Operating Characteristics (ROC) curve, and the F1-Score. The classification results indicated that important attributes had a substantial effect on accuracy, indicating that the models performed considerably better even with small amounts of data. Compared to models trained on the whole feature set, the reduced feature set considerably enhances classification accuracy while requiring less training time. This paper shows the relevance of feature selection in enhancing the performance of machine learning models for heart disease prediction.