In this project, we explored ECG (Electrocardiogram)-based authentication as a secure method for digital security, utilizing the unique and dynamic PQRST waveform of an individual’s heartbeat.
To start with, we introduced ECG and explained how it allows recording of electrical activity of the heart that forms a biometric signature that is quite unique compared to ordinary methods like fingerprints.
Then, we visualized ECG data from 90 participants and highlighted the PQRST segments to analyse heart activities in depth.

During the feature extraction stage, though initially we failed to use Fast Fourier Transform, we had a light bulb moment and recalled our high school mathematics and shifted our direction to Euclidean geometry. This approach permitted us to extract 56 distinct features from the ECG signals which largely simplified the model.

When we began training the model, we selected a K-Nearest Neighbours (KNN) classifier for supervised classification to obtain the highest accuracy ratio of 96% and a corresponding precision of 94%.
It is worthwhile to mention, that while the results were motivating, we also realized that some factors were obviously limiting. The model’s efficacy could get worsened with the variation of ECG signals caused by the patient’s physical and emotional conditions. Hence, we thought of possible options such as dynamic model training, reporting cases where breathing is abnormally quick for stress normalization, along with using a hybrid model of static and dynamic features.

In conclusion, we believe that ECG-based authentication holds great promise for enhancing digital security.