Childhood musical murmur classification with Support Vector Machine technique

Musical murmur, a typical occurrence of heart sound, frequently found in the pediatric population reflects no harm compared to murmur. 8 out of 10 children are then by nature prone to this physiological phenomenon. Correctly distinguishing musical murmur from pathological one is a critical assessment to avoid unnecessary treatment at vain cost disbursement and ease parents´ concern. This research investigates the algorithm development for automated high-yield classification with less time on analysis. 493 recorded heart sounds (HSs), comprised of 144 normal HSs, 141 musical murmurs, and 208 pathological murmurs, are first undergone through Preprocessing stage to form consistency of each incoming HSs. Derived HSs are extracted with Wavelet Packet (WP), as to form significant characteristics as a feature vector set. Then, Principal Component Analysis (PCA) is applied to the compactness of feature set in which the dimensionality reduction is attained through data-driven retrieval as to signify the most relevant features of HS record. Support Vector Machine (SVM) classifier trained by Radial Basis Function (RBF) kernel is constructed to detect and discriminate musical murmur from other murmur types by cross-validation process. Finally, classifier performance is evaluated by Confusion Matrix where an accuracy rate of 90.26% is ensured for the potential classification of model.