Electrocardiogram Signal Analysis Based on Statistical Approaches Using K-Nearest Neighbor

Abstract

AbstractThe performance of computer aided ECG analysis is very important for doctors and patients. Analyzing the ECG signal has been shown to be very useful in distinguishing patients from various disease. In this research, ECG signals is investigated using statistical approaches including Standard Deviation (SD), Coefficient of Variation (CoV), and Central Tendency Measure (CTM), and the machine learning model named, K-nearest neighbors (K-NN) model is used to identify them. With QRS complex extraction, the bit-to-bit interval (BBI) and instantaneous heart rate (IHR) were computed. CTM is measured IHR for ECG record of database. CTM highest value for IHR is detected for ten patients with normal rhythm with average value of 0.799 and low SD average of 5.633. On the other hand, the CTM for IHR of ten abnormal rhythm patients achived low value with average of 0.1645 and high SD average of 21.555. To validate the model, we utilized the standard MIT-BIH arrhythmia database. We used “twenty normal and abnormal rhythms” from the record of MIT-BIH database where each record is of one-hour duration. Experimental results proved that the proposed classifier named, K-nearest neighbor (K-NN) method gives the best accuracy 98.8%, best sensitivity 98.8%, and best specificity 89% for SD, best accuracy 98.2%, best sensitivity 98.23%, and best specificity 90% for CoV, and best accuracy 98.2%, best sensitivity 98.23%, and best specificity 90.2% for CTM respectively. Further, we distinguish between proposed model and state-of-art models including support vector machine and ensemble.KeywordsStatistical methodsECG signalK-NNCTMInstantaneous heart rate (IHR)