An Intellectual Framework for Heart Disease Detection Using Multi‐Scale Convolutional Autoencoder With Adaptive Gated Recurrent Unit

ABSTRACT Heart disease remains the leading cause of death worldwide. It is caused by a variety of unhealthy behaviors, including weight gain, high cholesterol, elevated triglyceride levels, hypertension, and more. These conditions impair the function of blood vessels, often leading to coronary artery disease, which is especially common among adults and the elderly. Incorrect diagnoses of heart disease can have devastating consequences, while early and accurate detection can help prevent life‐threatening events. To effectively manage patients at risk of heart attacks, it “is crucial to detect heart disease early and accurately.” Numerous studies have utilized various methodologies for classifying and predicting heart diseases. However, traditional algorithms are often inadequate for forecasting heart disease and cannot effectively handle high‐dimensional datasets. Furthermore, existing systems have not significantly improved the accuracy of heart disease diagnosis. In this study, we propose a deep learning‐based network to develop an advanced computerized system for heart disease detection. First, the required medical data is gathered from existing datasets. This data is then processed through an extraction and retrieval procedure. During this process, “deep features, Principal Component Analysis (PCA), and T‐distributed Stochastic Neighbor Embedding (TSNE) are extracted.” The resulting features are then input into a Multi‐scale Convolutional Autoencoder with Adaptive Gated Recurrent Unit (MCA‐AGRU) for heart disease recognition. To further increase the accuracy, the elements of the Gated Recurrent Unit (GRU) network are optimized using the Enhanced Fire Hawk Optimizer (EFHO) algorithm. Finally, the proposed approach is compared with traditional models to validate its effectiveness and superiority in heart disease recognition.