Classification of heart sounds using discrete and continuous wavelet transform and random forests

This study proposes an integrated approach to heart sound classification using wavelet analysis and random forests classifiers. The heart sounds were first segmented through detection of the S1 and S2 heart sounds using Shannon energy. Time- and frequency-based features derived from Discrete and Continuous Wavelet Transforms were used as feature vectors for the random forest classifier that categorized heart sounds into Normal, Murmur, Extrasystole, and Artifact. The segmentation process produced lower errors compared to related literature using the same dataset. In our study, the classification using features derived from DWT recorded the highest total precision for the noiser samples while CWT did the same for the less noisy samples. Overall, the proposed approach performed better than previous related studies.