**6. Conclusions**

The results we have obtained prove that the ECG signal is a valid biometric marker that is very robust to hardware variations, noise and artifacts presence, that is stable over time, and that is scalable over quite a solid number of users (>90). It is also hard to steal or mimic, is easy to measure, etc.

The biometric system allows for the achievement of a high operational speed, as just one heartbeat (average duration of less than 1 second) is enough to guarantee very good classification results (~90%). On the other hand, the outlier correction requires at least five heartbeats, which means that in a real-world application the overall response time will take at least 5 seconds.

The most promising algorithms for ECG identification are linear discriminant analysis (LDA), k-nearest neighbor (KNN), and neural networks (MLP). Another important conclusion clearly confirmed by our experiments is that PCA compression is not worth using at the data preprocessing stage, as in some cases it might reduce accuracy.

The following ideas might be interesting as potential future research topics: the estimation of system scalability for bigger datasets (e.g., mixed from different sources, and augmented using generative models), optimizing training hyperparameters for artificial neural networks, and performing a sequential analysis of neighboring heartbeats on the classification stage.

**Author Contributions:** M.P. has reviewed and edited this paper and he is the main corresponding author; Y.K. is the principle investigator who designed and performed the hardware/software implementation and experiments; V.K. presented the initial concept and has wrote article draft preparation.

**Funding:** This research received no external funding.

**Acknowledgments:** We thank Dmytro Sabodashko, PhD student at Lviv Polytechnic National University for assistance with collecting data for Lviv Biometric Data Set that was used in current research.

**Conflicts of Interest:** The authors declare no conflict of interest.
