Wavelet And Random Forest Based Arrhythmia Classification

Nowadays, cardiovascular disease is one of the harmful diseases to people's life and health. Lots of occurence cardiovascular disease companied by arrhythmia more or less. Long time of ECG monitoring would produce lots of data. The use of computer technology to preliminary screening and mark arrhythmia would greatly improve the efficiency of the doctors' diagnose, reduce the occurrence of misdiagnosis and missed diagnosis. Analysis on cardiac signal is focus on arrhythmia classification. Because individual difference, noise of signal, great variety of arrhythmia, lots of computation needed in process, some dispute existed on a few beats, this technology can not be well applied to clinic.This research use discrete wavelet and random forest to do arrhythmia classification. In terms of choosing dataset, we select some samples from MIT-BIH database as dataset. In terms of preprocess. we use discrete wavelet to remove high-frequency noise and baseline-wander. In terms of feature extraction, we use discrete wavelet, autocorrelation, PCA and variance and other mathematics methods, and extracted seven kinds of feature. The composed feature vector contains frequence-domain feature, time-domain feature and morphology feature.This research implemented arrhythmia classification system. And use experiment to test the influence of different parameters to accuracy, and do final comprehensive experiment to verify classification effect. mean accuracy of nine types of heart beat is 99.77%, mean sensitivity is 95.56%, mean specificity is 99.83%. The proposed feature extraction method is of certain guidance and reference in clinical arrhythmia automatic classification.