Cardioid graph based ECG biometric recognition incorporating physiological variability

This paper investigates ECG signal in different physiological conditions to identify different individuals. Data was acquired from 30 subjects, where each subject performed six types of physical activities namely walking, going upstairs, going downstairs, natural gait, lying with position changed and resting while watching TV. Then from the signals of these physiological conditions, specific features exclusive to each subject was extracted employing the Cardioid graph method. In this model, features were extracted solely from the graph derived using QRS complexes. Subjects were recognized with Multilayer Perceptron. Results were obtained through two approaches. In the former procedure, classification was performed on the whole dataset consisting of both training and testing set, which produced 95.3% of correctly classified instances. In the later approach the training and testing set was predefined where correctly classified instances were 93.9%. These results confirm that subject identification at different physiological conditions with Cardioid graph based technique produces better classification rates than previous study using only QRS complex.