A Machine Learning-Based System for Non-Invasive Multi-Vital Sign Monitoring Using PPG Signals

In this study, a non-invasive system capable of simultaneously monitoring five vital signs—body temperature, heart rate (HR), respiratory rate (RR), peripheral oxygen saturation (SpO₂), and blood pressure (BP)—using photoplethysmography (PPG) signals. To improve measurement accuracy, the system incorporates linear regression for body temperature calibration and specific value adjustments for SpO₂ estimation, achieving up to 99% accuracy. HR and RR signals are processed using third-order Butterworth filters. At the same time, BP is estimated using machine learning algorithms, with the Random Forest (RF) model achieving the best performance (MAE ±4.1 mmHg, RMSE ±5.02 mmHg). The system has been validated against standard clinical measurement tools, demonstrating consistent and accurate results across all parameters. These findings highlight the potential of the proposed system as a precise, non-invasive solution for real-time vital sign monitoring. Future development will focus on system integration, onboard data processing, and enhanced BP prediction through advanced feature extraction techniques.