Classification of Normal and Abnormal ECG Signals Using Machine Learning

Cardiac healthcare is one of the most crucial cares demanding field in any healthcare environment. A statistic by World Health Organization proves this statement with the conclusion that about 170.9 crores of people are losing their lives in a year because of the heart related issues. There are many works on understanding the deviations in heart’s electrical properties to pre-diagnosis the possibility of any heart related issues. In the presented work, an automatic rigorous method is proposed to classify and train the Electrocardiograph (ECG) signals such a way they can be used in predicting any possibility of deviation in the future. The study concentrates on three important aspects, the quality improvisation of the signal as the preprocessing part, identification of the best algorithm for classification and then the classification based on the prominent features. The ensemble empirical mode decomposition (CEEMD) method and temporal features of the signal are used for the preprocessing. Different machine learning algorithms are studied and the one with a high accuracy value is used in the classification. The Random Forest algorithm has customized entropy values with respect to the dynamicity of the heart related ailments. The conclusion shows promising results with better accuracy and also improved kappa values were obtained in validation.