Biometric person authentication framework using polynomial curve fitting-based ECG feature extraction

Abstract

The applications of modern biometric techniques for person identification systems rapidly increase for meetingthe rising security demands. The distinctive physiological characteristics are more correctly measurable and trustworthysince previous measurements are not appropriately made for physiological properties. While a variety of strategies havebeen enabled for identification, the electrocardiogram (ECG)-based approaches are popular and reliable techniques in thesenses of measurability, singularity, and universal awareness of heartbeat signals. This paper presents a new ECG-basedfeature extraction method for person identification using a huge amount of ECG recordings. First of all, 1800 heartbeatsfor each of the 36 subjects have been obtained from the widespread and large MIT-BIH database (MITDB) downloadedfrom the PhysioBank archive. Then the fiducial points of each heartbeat were determined and fourteen different featureswere extracted utilizing these fiducial points. Next, the polynomial curve fitting-based dimension reduction technique wasemployed on the extracted fourteen features. Furthermore, six celebrated classifiers including artificial neural networks(ANNs), decision trees (DTs), Fisher linear discriminant analysis (FLDA), K-nearest neighbors (K-NNs), naive Bayes(NB), and support vector machines (SVMs) were applied for the verification and performance evaluation of the proposedstudy. Also, as a different classifier, temporal classification and random forest was utilized for a benchmark classification.The highest performance was attained with 95.46% accuracy rate in the case of the SVM classifier. The experimentalresults emphasize that the proposed ECG-based feature extraction method gives insightful merit for biometric-basedperson authentication systems.