Abnormality detection in ECG using hybrid feature extraction approach

Biomedical signals like Electrocardiogram (ECG) contains essential information related to the functionality of heart. The pre analysis of ECG disturbances, aided by computer designed algorithms can prove to be efficient support in reducing cardiac emergencies. In this present method, dual tree complex wavelet transform (DTCWT) with linear discriminate analysis (LDA) also known as hybrid feature extraction are employed for denoising and dimensionally reduced non linear feature extraction respectively. The classification and analysis of ECG dataset into normal and abnormal beats is done by independently deploying five classifiers like support vector machine (SVM), decision tree (DT), back propagation neural network (BPNN), feed forward neural network (FNNN) and K nearest neighbour (KNN). The outcomes of proposed work are compared with pre existing methods. The highest percentage accuracy of 99.7% is achieved using BPNN, SVM and KNN. The simulation results show that the shift invariance nature of DTCWT provides a robust technique for non linear and non stationary ECG signals.