Precision blood pressure prediction leveraging Photoplethysmograph signals using Support Vector Regression

To facilitate the operation of more sophisticated medical robots, blood pressure prediction technology was developed using Photoplethysmograph (PPG) signals from a single finger, using the Support Vector Regression (SVR) method. The data collection process involved 110 participants aged 20 to 70 years for modeling and validation. The model training phase was carried out with various parameter variations to obtain the optimal model based on the Mean Absolute Error (MAE) value. The blood pressure estimation results showed an average error of around 2.78 mmHg for systolic pressure and 7.34 mmHg for diastolic pressure. Validation on 30 new participants revealed a slight increase in the average error, which was around 4.23 mmHg (with 93.90 % accuracy) for systolic pressure and 5.12 mmHg (with 96.64 % accuracy) for diastolic pressure. These results, which are characterized by a low error rate, indicate that the SVR model is able to predict blood pressure accurately and consistently, both on training data and new data that was previously unseen.