Identification and diagnosis of chronic heart disease: A deep learning-based hybrid approach
Abstract
Chronic heart disease has emerged as a challenging issue in the healthcare sector that needs serious attention to save the lives of millions of cardiac patients. The precise diagnosis of this disease in the early stages can reduce the devastating effect it has on human life. To address this issue, this study proposes a hybrid deep learning (DL)-based approach that combines two versatile DL models, namely, bidirectional long-short-term memory (BLSTM) and bidirectional gated recurrent unit (BGRU), resulting in an efficient hybrid DL model named BLSTM-BGRU. The BLSTM part captures long-term relationships between dataset attributes, guaranteeing the preservation of the patient’s historical data, which is essential for forecasting the patient’s health conditions. The BGRU part improves the computing efficiency of the model by lowering the number of trainable parameters and reducing the effect of vanishing gradient problems. The integration of BLSTM and BGRU helps the model to learn the short-term variations and long-range dependencies in heart disease attributes such as heart rate, respiratory rate, etc. The proposed model captures contextual dependency in forward and backward directions, resulting in improved heart disease diagnostic accuracy by learning long-range relationships between attributes and complex sequences. To determine the efficiency of the BLSTM-BGRU model, the MIT-BIH dataset, which consists of five different types of ECG signals, was used. The dataset consists of more normal class samples than the rest of the four classes. Therefore, we used the SMOTE dataset balancing technique to balance the dataset, thereby avoiding the model overfitting problem and improving its efficiency. Alongside the proposed model, we also investigated the performance of four other of the most versatile DL models on both unbalanced and balanced datasets. The proposed model achieved training and testing accuracy of 99.90% and 99.58% on an unbalanced dataset and 99.95% and 99.70%, respectively, on a balanced dataset. The results highlight the importance of the proposed BLSTM-BGRU model using both balanced and unbalanced datasets, showing its significance and versatility for the identification of heart disease, resulting in enhanced heart disease prevention and management.