Evaluation of characteristic frequency features in healthy and diseased ECG via k-NN classifier

This paper presents an evaluation of characteristic frequency features in healthy and diseased ECG via k-NN classifier. Initially, a total of 264 segment samples are obtained for healthy, bundle branch blocks, dysrhythmia cardiomyopathy conditions from the PTB Diagnostic ECG database. The signal is preprocessed to obtain the power spectral density. The characteristic frequency for each segment sample is then extracted. Six distinct characteristic frequencies have been observed with varying pattern of power distribution in each ECG condition. Power related to specific characteristic frequencies is then successfully implemented for feature classification via k-NN with 100% accuracy during training and testing. Reliability of the characteristic frequencies as ECG descriptors has also been confirmed via k-fold cross-validation.