Improvement in ECG based Biometric Systems using Wavelet Packet Decomposition (WPD) Algorithm

Abstract

In this thesis, a non- ducial approach based on wavelet packet decomposition (WPD) algorithm for repeated examination of solitary lead electrocardiogram (ECG) for individual identi cation is planned and tested. Multiple samples of ECG wave are extracted considering R-peak as a reference and WPD algorithm is applied for feature extraction. This feature le is fed as an input to a machine learning classi er i.e. random forest in order to classify the individuals. In this work, records from publicly available MIT/BIH arrhythmia dataset have been utilized to evaluate the proposed system. Best result relies on third level of wavelet decomposition using Daubechies wavelet to analyze the signal. Furthermore, ranker search method is used in conjunction with relief attribute evaluator for feature selection and random forest classi er is applied by generating 100 trees. It is shown that the method is e ective for quantifying the classi cation of arrhythmia ECG signals with accuracy of 92.62 %.