Heartbeat Abnormality Detection in Phonocardiogram Signals using Wavelet Time Scattering and Optimized KNN Classification

Heart auscultation continues to play an essential role in heart health diagnosis. However, many places worldwide have a shortage of suitably qualified medical practitioner's adept at this ability. This highlights the critical need to develop accurate automated systems for evaluating Phonocardiogram (PCG) data. PCGs are acoustic recordings that capture the noises made by the heart during its systolic and diastolic cycles. To solve this issue, we suggest using Wavelet Time Scattering with an optimized XGBoost classifier and K-Nearest Neighbors (KNN) classifier to detect irregular heartbeats in PCG signals. The results are promising, as the optimized KNN classifier obtains an impressive accuracy rate of 92.5% when combined with five-fold cross-validation, which is better than XGBoost classifier, which gains 87.93%. This demonstrates the efficacy of the optimized KNN in improving the automated interpretation of PCG data and assisting in the early diagnosis of heart-related problems.