Application of neural network for ECG-based biometrics system using QRS features

Applications of Biometrics technology are extremely popular today, ranging from access control to automation. Fingerprint is the oldest and the most widely used biometrics technology. However, its key features are externally exposed which make it tend to be easily forged. This study investigates the possibility of electrocardiogram (ECG) signal as an alternative modality for biometrics systems. Besides that, the study is conducted using the ECG database under arrhythmia conditions to accommodate the real-world application since arrhythmia exists in large-scale world populations. In this study, a total of 8,972 datasets from 47 subjects were modeled using a machine learning technique (i.e., one-dimensional convolution neural network or 1-D CNN). The results showed that the accuracy (F1-score) of 92% and 0.25 of loss was achieved. Furthermore, we prove that the proposed model is a good fitting based on the visualization plot of the train-test. These findings show that the proposed model is reasonable enough for an ECG-based biometrics system though it's not the best in the literature.