Arrhythmias discrimination based on fractional order system and KNN classifier

Abstract

The electrocardiogram (ECG) signal may contain useful information about the nature of the diseases afflicting the heart. However, the cardiac abnormalities information cannot be easily and directly monitored by the human eye for large amount of ECG data. Hence, computer assisted methods are very important to monitor cardiac health easily and accurately. In this paper, ECG Normal beats (N), Premature Ventricular Contraction beats (PVC) and Bundle Branch Block beats (BBB) (right and left) discrimination through a new modeling technique of the QRS complex frequency content is investigated. Because of the fractional slope behavior of the power spectrum of the QRS complexes, the proposed model of the QRS complex frequency content is a linear fractional system of commensurate order. These fractional models of normal and abnormal QRS complexes are used to extract useful information about the functional activity of the heart. In this context, the features used for the ECG beats discrimination are the parameters of the linear fractional system of commensurate order. These pertinent features are then classified using the K-Nearest Neighbors (KNN) classifier, because is simple and more suitable for this type of features. The performance and the effectiveness of the proposed method is evaluated and validated on the ECG signals of the MIT/BIH arrhythmia database. The proposed method has achieved 95.161% of accuracy for 23 records form the database.