Performance analysis of feature selection methods for feature extracted PCG signals

Analyzing and characterizing the Phonocardiogram (PCG) signal is important for Diagnosing the valvular Heart Disease. The PCG can record heart sounds, noise and the additional sounds. Since PCG is recording the heart sound, it is important to analyze the clear PCG input signal only. The analyzation of the PCG signal will be consisting of segmenting the signal into S1 and S2 and then compare, whether the PCG is normal or abnormal. In the existing system the wavelet decomposition approach is used to analyze the PCG signal. Features are extracted from a PCG signal in frequency domain to classify signals. In the proposed approach the Feature selection reduces features provided for classification. Coiflet is used for feature extraction, and different feature selection Statistical methods are used. Information Gain (IG), Mutual Information (MI) etc. Feature selection methods are compared using classifiers like kNN, Naïve Bayes, C4.5, and SVMs. In this paper, two methods are used to analyze the PCG and the accuracy of the Information Gain (IG) is improved when compared to Mutual Information (MI).