Nearest neighbor search with locally weighted linear regression for heartbeat classification

Automatic interpretation of electrocardiograms provides a noninvasive and inexpensive technique for analyzing the heart activity of patients with a range of cardiac conditions. We propose a method that combines locally weighted linear regression with nearest neighbor search for heartbeat detection and classification in the management of non-life-threatening arrhythmia. In the proposed method, heartbeats are detected and their features are found using the Pan–Tompkins algorithm; then, they are classified by locally weighted linear regression on their nearest neighbors in a training set. The results of evaluation on data from the MIT-BIH arrhythmia database indicate that the proposed method has a sensitivity of 93.68 %, a positive predictive value of 96.62 %, and an accuracy of 98.07 % for type-oriented evaluation; and a sensitivity of 74.15 %, a positive predictive value of 72.5 %, and an accuracy of 88.69 % for patient-oriented evaluation. These results are comparable to those from existing search schemes and contribute to the systematic design of automatic heartbeat classification systems for clinical decision support.