Real-Time Blood Pressure Change Classification Using Enhanced PPG Signals and Convolutional Neural Networks

Photoplethysmography (PPG) is widely used as a non-invasive and cost-effective technique for monitoring cardiovascular activity and assessing blood pressure (BP) variations. However, PPG signals are often affected by noise and motion artifacts, which can reduce signal reliability and negatively impact clinical interpretation and machine learning performance. This study proposes a signal enhancement and classification framework to improve the accuracy of BP change classification using PPG signals. The proposed approach enhances signal quality by reducing noise while preserving important physiological waveform characteristics, enabling more reliable feature extraction. The enhanced signals are then utilized as input to a Convolutional Neural Network (CNN) to learn discriminative temporal and morphological features associated with BP variations. Experimental results demonstrate that the proposed framework achieves a training accuracy of 96.66% and a validation accuracy of 95.31%, outperforming conventional preprocessing approaches such as hard and soft thresholding. These findings highlight the potential of integrating adaptive signal enhancement with deep learning techniques to improve the robustness and reliability of non-invasive BP monitoring systems. The proposed framework offers promising applications for real-time and clinically relevant cardiovascular monitoring.