Non-invasive Machine Learning approach for classifying Blood Pressure using PPG Signals in COVID situation

ABSTRACT

Blood pressure is one of the possible factors that cause cardiovascular diseases. It is one of the useful parameters for early detection, using which we can diagnose and treat cardiac diseases. Continuous monitoring of blood pressure can help us to maintain good health and to have a longer life span. At present, BP estimation is principally based on cuff-based techniques[1] which can cause inconvenience or discomfort to patients. ECG is one of the cuff-based methods to estimate or classify Blood Pressure. Nowadays, Studies are taking place on non-invasive and cuff-less-based methods and one of them is PPG signals (photoplethysmography). PPG is a non-invasive optical method for estimating the blood volume changes per pulse[21]. We can also say that the PPG signal indicates the mechanical activity of the heart[8]. In this paper, we proposed a non-invasive method using a whole-based approach that uses raw values from PPG signals to classify blood pressure. Using Machine learning algorithms to classify blood pressure is a feasible way for the analysis and predicting the results. In this paper, we applied various machine learning models(Random forest, Gradient boost, and XGBoost). In order to avoid overfitting, we used Repeated-stratified k-fold cross-validation and obtained enough accuracy in classifying the BP. when compared to the parameter-based method, our method(whole based method) is independent of the PPG waveform of a signal. © 2021 IEEE.