Heart CAP

Medical diagnostics is an ever-evolving domain. In recent times, heart diseases have spread rapidly in younger and older people. Thus, an urgent demand is to identify and classify heart illnesses accurately. Traditional invasive methods are available for detecting heart-related diseases. However, they are costly. Thus, in this investigation, we aim to diagnose heart diseases using modern data-driven and machine-learning techniques. Specifically, this work compares traditional and novel feature engineering techniques for heart disease classification. The first step in this study is to prepare the dataset carefully. Data preparation involves dealing with missing and categorical values. In the data preparation, we used feature scaling to ensure uniformity among features. Then, the model is developed and trained using the training data and tested using the testing dataset. A well- designed encoder and decoder-based Undercomplete Autoencoder (UAE) architecture is built for feature engineering. The Principal Component Analysis (PCA) technique is used for traditional feature engineering. The logistic regression model is trained using the data that has been AE-encoded or PCA-transformed. We have taken ten features for the heart disease classification task in both cases. The accuracy, precision, recall, F1-score, and receiver operating characteristic - area under the curve (ROC AUC) measures are considered to evaluate the performance of our model. The results demonstrate that the model identifies heart diseases effectively. The UAE outperforms PCA on the Cleveland Heart Disease Dataset. While PCA features are effective, the features of AE can help identify potential cases of heart disease with greater efficacy. The important features of the classification decisions according to each method were also described in the study. Thus, this contribution will pave the way for more accurate medical diagnosis.