Frequency Rank Order Statistic with Unknown Neural Network for ECG Identification System

Abstract

Electrocardiograms (ECG) contain biological information which is unique to the individual. In this paper, an ECG identification system, which uses Frequency Rank Order Statistics (FROS) as a feature extraction method and Back-Propagation Neural Network (BPNN) classifiers to identify 'other classes', is proposed. FROS handle different ECG states and BPNN classifiers, with random input weights, are used to generate a relatively high accuracy model for the identification system. Additionally, in the output layer, classified patterns are categorized according to the maximum value of the output layer nodes. Similar data is grouped into one category for the final identification result. Experiments show that the BPNN classifier produces more accurate results than an SVM and Bayesian classifier achieve on average. The proposed approach also out-performs SVMNN and LVQNN. The identification system, put forward in this paper, may be applied to an intelligent vehicular system, as an application example.