Selection of mother Wavelet to characterize 5 kinds of heartbeats using the discrete Wavelet transform

Abstract

It’s documented the review of a methodology to select those mother wavelet that allow to characterize heartbeats by a best way, from the analysis of the coefficients of approximations and details, obtained by applying the discrete wavelet transform to 5 kinds of heartbeat. The indicator which allows the mother wave-lets identification is based on analysis of concentrated energy in approximations and details, of the first four levels of decomposition. The generated feature vectors were entered into Support Vector Machine with Poly-nomial kernel and radial basis function, allowing quantifying the characterization capability of mother wave-lets that achieved interesting energy indicators, from the results of classification. Fewer errors were obtained using the Reverse Biorthogonal mother wavelet of order 3.1 (rbio3.1), which is not considered commonly in research papers, with a support vector machine with polynomial kernel, whose percentage error was 2.71%.