Optimized machine learning for real-time, non-invasive blood pressure monitoring

Continuous, non-invasive blood pressure (BP) monitoring is essential for managing cardiovascular health; yet, traditional cuff-based systems are unsuitable for real-time use. This study presents a hybrid machine learning framework that fuses photoplethysmography and electrocardiography signals with optimized feature selection to achieve accurate and efficient BP estimation. By leveraging complementary signal characteristics, key physiological features such as pulse transit time and pulse arrival time are extracted to improve predictive performance. A Random Forest model tailored for physiological data yields state-of-the-art accuracy, with a mean absolute error of 1.602 mmHg for diastolic and 2.082 mmHg for systolic BP, exceeding clinical standards. The model maintains high accuracy while reducing feature dimensionality by 80%, supporting real-time deployment on wearable devices. This AI-driven approach addresses sensor variability and computational constraints, offering a practical solution for continuous, cuffless BP monitoring in personalized and digital healthcare.