<p>Heart sound classification is essential in cardiovascular diagnostics by identifying normal and pathological patterns within phonocardiogram (PCG) signals. Due to their non-stationary nature and the presence of both physiological and pathological variations, accurately classifying heart sounds presents significant challenges. This study introduces a comprehensive framework for classifying heart sound data via the integration of three different feature extraction techniques and three different learning techniques. The pipeline begins with capturing raw heart sound, which undergoes a series of preprocessing steps including resampling, bandpass-filtering, denoising, segmentation, wavelet analysis, continuous wavelet transform, scattering wavelet transform and MFCC to enhance the quality and extract relevant features. Following preprocessing, the processed signals are fed into various classification models such as SVM, CNN, and CRNN to identify and categorise heart sound patterns. The proposed method is based on a ensemble-learning approach that utilises KNN as a ensemble-classifier, combining the predictions of multiple base-level classifiers, such as SVM, CNN, and CRNN, to make a final prediction. The ultimate goal is to provide an accurate classification result, which may aid in the detection of cardiovascular disorders in their early stages. This design showcases the possibility of enhancing the accuracy and reliability of cardiac sound classification systems via the integration of conventional signal processing techniques with cutting-edge deep learning methodologies.</p>