Motion Artifact Classification in PPG Signals Using 1D CNN and Active Learning

Photoplethysmography (PPG) signals are widely used in clinical and wearable healthcare systems to monitor vital signs such as heart rate, oxygen saturation, and blood pressure. However, their susceptibility to motion artifacts (MA) often compromises signal quality and leads to inaccurate readings. Traditional approaches to MA mitigation typically require additional sensors or complex signal processing, increasing cost and system complexity. In this work, we present a methodology for the classification of motion artifacts in PPG signals using a 1D Convolutional Neural Network (1D-CNN), combined with a targeted active learning strategy to enhance model generalization. The model is trained on manually labeled 2-second signal frames and iteratively improved using uncertainty sampling over an external dataset. Experimental results show that the 1D-CNN achieves performance comparable to a decision tree classifier based on handcrafted features, while offering key advantages such as end-to-end learning from raw data. More importantly, the integration of active learning leads to consistent performance improvements and reduces the need for extensive manual annotation. These findings support the use of deep learning and active learning as a practical and scalable solution for detecting motion artifacts in PPG signals, without relying on additional hardware or extensive preprocessing.