Hearing the heart beating… for data science
A borrow idea
Looking for interesting ideas for a project, I found an electrocardiogram (ECG) database created and curated to learn if the ECG signal could be used as a biometric human identification. I took the challenge, proposing an algorithm to identify a subject just by knowing general demographic information and the ECG signal.
But why the ECG signal?
The ECG signal represents the electric activity of the heart and, although it is similar for all humans, this biological signal depends on the anatomy of each person, meaning it is too specific, or in few words, the ECG is personal. Besides, a simple ECG signal with a general good quality can be obtained currently by wearable devices connected to the cloud, making easy to associate the personal ECG to web based services, for example.
The Power of Features
But how can the ECG signal be used for identified a subject? Characteristics of the ECG signal are a first option. But then, how these characteristics should be selected? And which characteristics are better for the purpose of biometric identification? Should ECG characteristics be combined with other information? How many features are too many (or too few) features? It became clear that selecting the proper characteristics of the ECG signal was key to obtain a good classification.
It was at least clear that this was a classification task, and machine learning could provide a good option up to the challenge. So, after the ECG characteristics were selected, a model was selected, trained and tested, and then more features included, then removed, and after each model training, the performance of the model was evaluated. These cycles led to an optimization of the model, and a good grasp on the possibility to used the ECG as a biometric identification method.
The idea to use the ECG as a biometric identification due to its uniqueness for each individual showed potential. With the possibility of ECG acquisition by wearable devices connected to the cloud, the access to common web based services could be ensured by this additional security level, even if there are some additional challenges to take into account, such as how aging or disease can impact the ECG signal, but that’s a challenge for the next post.
Check the complete project at: https://github.com/franciscoj-londonoh/ECG-based-Biometric-Identification