Abstract
BACKGROUND: Cardiovascular diseases (CVDs) represent the leading cause of mortality globally, accounting for more than 17.9 million fatalities annually. The prompt and precise forecasting of cardiovascular diseases is crucial for timely management and enhancing patient outcomes. Analysis of heartbeat signals is crucial for the early identification of cardiovascular diseases. METHODS: This research presents a resilient neural network architecture for the automated classification of heartbeat signals by integrating Convolutional Neural Networks (CNNs) with the Long Short-Term Memory (LSTM) form of Recurrent Neural Networks (RNNs). We implemented comprehensive preprocessing approaches, such as data balance, normalization, augmentation, and feature extraction, to improve the quality of the heartbeat signals and utilized publicly accessible datasets to test our model. RESULTS: On the public datasets used, the model achieved up to 96.04% test accuracy for binary classification and 97.83% for three-class classification, indicating competitive performance against recent baselines evaluated on comparable settings. CONCLUSION: The proposed audio-based diagnostic framework demonstrates strong predictive performance for heart sound classification, revealing the feasibility of incorporating auscultation signals into automated decision-support systems. Future research will enhance the model, augment the dataset, and investigate supplementary predictive variables to improve its precision and clinical relevance.