CardioPRINT: Biometric identification based on the individual characteristics derived from the cardiogram

Abstract

ObjectiveThis paper investigates the potential of cardiogram-derived traits from electrocardiogram (ECG) and impedance cardiogram (ICG) for biometric identification. Additionally, the influence of induced emotions on cardiogram attributes and their impact on identification accuracy is explored. MethodWe compare 7 machine learning classifiers using a dataset gathered from 202 individuals to identify the highest-performing classifiers. Subsequently, we analyze three different feature sets employing (ECG-only, ICG-only, and both ECG and ICG). Additionally, we investigate the performance of classifiers under altered emotional states to assess classifiers’ robustness. ResultsThe analysis demonstrates that models employed with both ECG and ICG have the highest statistically significant accuracies. The best-performing Random Forest (RF) model using both ECG and ICG achieves an average accuracy of 97.2 %. All models reveal a decrease in classification accuracies (∼13 %) when not trained and tested under identical emotional conditions. ConclusionOur findings suggest that integration of ECG and ICG-based features could increase the accuracy of identification compared to a single-signal-based approach. Although certain models show slight robustness to altered emotional states, the effect of the emotion is evident and future selection of cardiogram-based features, as well as biometric models, should consider emotional responses.