CADFRA: Coronary Artery Disease Feature Reduction with Autoencoder for Optimistic and Effective Classification

Heart or Coronary Artery Disease (CAD) is a global health concern. It is necessary to detect it early for an improved healthcare results. There are wide range of detection mechanisms such as angiography. It is frequently used to forecast cardiac diseases. However it is costly and requires skilled technician. Alternative to angiography, there is a necessary for developing the cost-effective, non-invasive methods. Such methods should reduce financial burdens and boost diagnostic precision. Machine learning techniques provide promising options to costly methods. However, feature extraction and reduction techniques have been overlooked by the researchers. This article explores the potential of machine learning approaches, focusing on data preparation and reduction techniques to enhance diagnostic performance. Specifically, we employ the under-complete Autoencoder to extract latent features and reduce complexity. This is leading to improved classification performance. The proposed work is evaluated on six machine-learning models. The features extracted achieving a consistent AUC-ROC score of 95%. The highest accuracy of 95% on the Cleveland dataset. Our proposed machine learning-based Coronary Artery Disease Feature Reduction with Autoencoder (CADFRA) model showcases optimistic and competitive outcomes compared to state-of-the-art techniques.