Biometric Authentication via Electrocardiogram Traces

With the heightened awareness of security measures, biometric methods such as facial recognition and fingerprinting that rely on physiological characteristics of a particular user have become increasingly important for authentication. Previous research has shown that machine learning techniques can be used to obtain viable biometric markers from electrocardiograms (ECGs). As biometric methods like these begin to be applied at a wider scale and on a wider range of devices, finding efficient representations and algorithms becomes a growing concern. In this study, we aim to observe the tradeoffs between accuracy and computational efficiency in classifiers with varying levels of complexity. Using the PlexNet ensemble as a basis, we examine the key elements of its architecture through individual performance of its components; we also develop and test a smaller fusion of these elements. Our findings indicate that higher complexity ECG classifiers do not guarantee the most efficient performance compared to corresponding base models.