Abstract
The classification of Phonocardiogram (PCG) time series, which is often used to indicate the heart conditions through a high-fidelity sound recording, is an important aspect in diagnosing heart-related medical conditions, particularly on canines. Both the size of the PCG time series and the irregularities featured within them render this classification process very challenging. In classifying PCG time series, motif-based approaches are considered to be very viable approach. The central idea behind motif-based approaches is to identify reoccurring subsequences (which are referred to as motifs) to build a classification model. However, this approach becomes challenging with large time series where the resource requirements for adopting motif-based approaches are very intensive. This paper proposes a novel two-layer PCG segmentation technique, called as PCGseg, that reduces the overall size of the time series, thus reducing the required for generating motifs. The evaluation results are encouraging and shows that the proposed approach reduces the generation time by a factor of six, without adversely affecting classification accuracy.