Automated Detection of Chronic Heart Failure from Heart Sound Signals Using Integrated Learning Models

Chronic Heart Failure (CHF) is a lifethreatening condition that requires timely and accurate diagnosis to reduce mortality rates. This work presents a hybrid approach that combines classical machine learning and end-to-end deep learning techniques for detecting CHF from heart sound recordings. The study utilizes the PhysioNet heart sound dataset, which includes phonocardiogram (PCG) signals and audio recordings labelled as normal or abnormal. Initially, systolic and diastolic features are extracted from the PCG signals and used to train a Random Forest-based classical machine learning model. In parallel, raw heart sound recordings are directly processed using a deep learning model to capture complex patterns. To further enhance prediction performance, features obtained from both models are aggregated and used to train a final recordinglevel classifier. This integrated approach leverages the strengths of both feature-based and deep learning methods. Experimental results demonstrate that the hybrid recording model achieves higher accuracy compared to individual models. The system is capable of classifying unseen heart sound recordings effectively as normal or abnormal. Overall, the proposed method offers a reliable and efficient solution for early detection of chronic heart failure, especially in scenarios where expert medical analysis may not be readily available.