Personal Identification Based on Smartwatch

This project tried to find a new way for personal identification based on the sensor we have on the smartwatch. As today’s smartwarch can easily detect raw PPG(Photoplethysmography) and ECG(Electrocardiography) signal, we tried to use machine learning algorithm to see if these two signals can give us enough information for personal identification.

We cleaned raw PPG and ECG data and extracted 20 different features from filtered signals using Python. Then, we built random forest models to identify different people using selected features. We also tried an LSTM model with Keras to conduct analysis.

The results are pretty promising. It seems 3 HRV features and 4 spatial features can provide most information for personal identification. LSTM does provide a pretty good result, but we believed that, with more data, it can have a much accurate result.

Here is a video introduction to our project.