Wavelet transform and support vector machines for the arrhythmia identification

In this paper we present several characteristics extraction schemes set of both normal beatings and those with four different types of cardiac arrhythmias through the wavelet transform. One of these group schemes, randomly chosen, was used to determine the optimal parameters of the Polynomial Kernel (C & D), and the radial basis kernel (C & gamma) from which we could obtain the best classification results for the Support Vector Machines. According to these parameters we applied different classification tests with all the training set, which led us to an error validation percentage of 2.93 with a support vector machine with a radial basis kernel, 11.72 with a Bayesian Classifier, and 3.63 with a multilayer perceptron. With the purpose of reducing the error validation percentage, we applied the principal components analysis technique for both the effective selection of characteristics and the reduction of the dimensionality of the characteristic vectors on each training set. As a result, we could observe that the support vector machines evaluated with these new training set were the most consistent, having a lower error validation percentage of 1.93; and the Bayesian classifier improved its classification ability, while the multilayer perceptron did not have an accurate response to the effective selection of characteristics.