Machine learning based improved automatic diagnosis of cardiac disorder

Heart diseases are one of the most common diseases these days. The common cardiovascular diseases are usually being diagnosed by the manual stethoscope by doctor. In many developing countries doctors are not available in primary health care centers in rural areas. This paper proposes a method to diagnose and detect the abnormal heart frequencies using discriminatory features of the heart sound by machine learning. Mel frequency cepstral coefficients study has been done to excerpt the features from the heart sound, which increases the sensitivity of the results. Support Vector Machine is used to train and test the features extracted. The proposed method uses the Butterworth filter for pre-processing of noise removal to clean the signal. Time complexity has been decreased and due to the logic implemented the device database can get update by itself as far as it gets in use and doesn't need human intervention making it completely automatic. The proposed method is tested on a comprehensive database of heart sounds with different non-overlapping testing sets. The proposed method achieved the best accuracy of 95 % during classification process. The experiment results indicate that the proposed method is efficient for classification of healthy/unhealthy heart sounds and computationally cheap making it suitable for real time applications.