Machine Learning Techniques for the Cuffless Estimation of Blood Pressure using PPG Signals MATLAB

Abstract

In this task we are comparing different Machine learning approaches for the cuffless estimation of blood pressure (SBP and DBP) using PPG signal. First, we extract the features from the dataset, to train the model we are using the 219 patients datasets from that three dataset were largely comipted.process of feature extraction are as follows:

• importing PPG signal
• filter signal using Chebyshev Type-II filter
• Normalize filtered signal
• Extract ST, DT, Width 1 (50%) and Width 2 (66%)
• Storing the Extracted Feature data on Dataset.x1sx file

Preparing the training data by joining the Extracted Feature with the dataset contains the patient information such as Age, Height, Weight, SBP, DBP, BMI. Now we train the Regression model and compare the model which gives the least MAE. In this task we are using Gaussian Process Regression, Linear Regression, Support Vector Regression, Artificial Neural Network, Bagged Tree and Boosted Tree. The MATLAB Script with the result are shown in Video.

By visualizing the above graph of MAE for both the model we can see that SVR model gives the lesser MAE amongst all the other Machine learning approaches, so we can considered it as the best model with MAE of 13.53 with standard deviation of 0.13 mmHg from the performance of the code we can see that preprocessing the data takes around 42 seconds, for preparation of the training data takes 0.3 seconds and to train the six SBP Model it takes 304 seconds wherea to train DBP model it takes 645 seconds. From the result we can see tell tat SVR model gives the best result for the estimation of SBP and DBP.