Binary Classification of Heart Failures Using k-NN with Various Distance Metrics

Magnetocardiography is a sensitive technique of measuring low magnetic fields generated by heart functioning, which is used for diagnostics of large number of cardiovascular diseases. In this paper, k-nearest neighbor (k-NN) technique is used for binary classification of myocardium current density distribution maps (CDDM) from patients with negative T-peak, male and female patients with microvessels (diffuse) abnormalities and sportsmen, which are compared with normal control subjects. Number of neighbors for k-NN classifier was selected to obtain highest classification characteristics. Specificity, accuracy, precision and sensitivity of classification as functions of number of neighbors in k-NN are obtained for classification with several distance measures: Mahalanobis, Cityblock, Eucleadian and Chebyshev. Increase of the accuracy of classification for all groups up to 10% was obtained using Cityblock distance metric in binary k-NN classifier with 19 - 27 neighbors, comparing to other metrics. Obtained results are acceptable for further patient’s state evaluation.