Approach to cascade classifiers for identifying heart-beats

Abstract

This work describes the using of cascaded classifiers to identify heart-beat patterns. These patterns belong to classes no considered during training. We employed supervised learning machines such as support vector machines (SVM) and multilayer perceptron (MLP). The cascaded classifiers were validated with 5 different kinds of heart-beats. The discrete wavelet transform (DWT) was used for feature extraction. For each decomposition level, only the 4 largest coefficients were taken from approximations and details. The DWT uses 6 decomposition levels and Daubechies-4 mother wavelet. The achieved classification error was 3,55%.