Diagnostic accuracy enhancement for cardiovascular disease prediction using dual optimized feature selection and fuzzy-based deep learning model

Abstract

The most frequent reason for mortality in many nations is cardiovascular disease. Past experience and current clinical testing of diagnosing patients with comparable symptoms are frequently used by doctors to make the diagnosis of cardiovascular disease. Heart disease patients need to be diagnosed as soon as possible, treated as soon as possible, and closely monitored. Numerous data mining techniques have already been employed to diagnose and forecast heart conditions in order to meet their objectives. To help doctors forecast and detect cardiovascular disease, deep learning and machine learning may provide a stronger basis for prediction and decision-making from healthcare data sectors around the world. The aim of the research is to propose an accurate algorithm for the prior prediction of heart disease using dual feature selection methodologies. The features are selected by utilizing feature selection methods such as LASSO and MR-MR. The early prediction of cardiovascular disease (CVD) is performed using an improved fuzzy-based TabNet deep learning model with a fuzzy foundation. The dataset is considered from the Kaggle Heart Disease Repository. The area under the curve (AUC) for the recursive operating characteristic curve is estimated for the proposed algorithm. Additionally, error measures like mean absolute error (MAE), mean squared error (MSE), and root mean squared error (RMSE) are estimated for the model's predictions, and the magnitude of MSE is 0.038, MAE is 0.180, and RMSE is 0.195, respectively. The best approach for classifying and predicting CVD is the integration of the enhanced TabNet algorithm and fuzzy foundation. The suggested approach lowers costs and improves medical care for predicting heart illness. The strength of the suggested model is relatively satisfying, and it reveals good accuracy in predicting indications of heart disease in a specific individual when compared with previously implemented classifiers.