Noninvasive Blood Pressure Classification Based on Photoplethysmography Using Machine Learning Techniques

Blood pressure (BP) is one of the four vital signs that offer crucial medical insights into cardiovascular activity. High BP is associated with an increased risk of diseases such as heart attacks and strokes. Traditional BP measurement methods, including invasive and cuff-based devices, have limitations in providing continuous monitoring and can be uncomfortable for individuals. In contrast, wearable devices offer a promising solution for ambulatory care and public health monitoring by enabling frequent BP measurements in nonclinical environments. To meet this requirement, we propose an approach for cuff-less and continuous BP classification using photoplethysmograph (PPG) signals and machine learning (ML) techniques. PPG is a light-based method used to detect variations in blood volume with each heartbeat, offering a noninvasive approach for evaluation. This technology is costeffective, accessible, and allows for continuous usage. In this research, PPG signals collected from various individuals were subjected to preprocessing and feature extraction. To enhance the performance of ML algorithms and address concerns related to computational complexity and overfitting, feature selection techniques (three techniques) were implemented. These techniques aimed to strategically choose relevant features and subsequently train and evaluate the ML algorithms (eight algorithms) using these selected features. The output of the algorithms is in the form of four classes: normotension (NT), prehypertension (PHT), stage 1 hypertension (S1HT), and stage 2 hypertension (S2HT). The light gradient boosting machine (LightGBM) algorithm, combined with the SelectFromModel feature selection technique, achieved the highest performance, boasting an accuracy of $84.61{{\% }}$ on the test data.