Biophysics and Beyond

The post-conflict reconstruction of Karabakh presents an urgent public health challenge: more than 10,000 displaced individuals had returned to liberated territories as of January 2025, with official targets of 140,000 returnees by end of 2026, substantially outpacing healthcare infrastructure development. This paper presents a conceptual framework for deploying a machine learning-enhanced, fingernail-mounted photoplethysmography (PPG) device as a low-cost, non-invasive vital sign monitoring tool in these underserved communities. Building on a previously developed and TÜBİTAK-funded prototype, we propose an upgraded architecture integrating Random Forest-based artifact detection, LSTM-based adaptive noise cancellation and autoencoder-based anomaly detection. As a limited preliminary computational feasibility test, the proposed Stage 1 artifact detection workflow was evaluated on the open-access Pulse Transit Time PPG Dataset (n=3 subjects, 108 windows) under Leave-One-Subject-Out cross-validation. PPG-only features achieved 36.1% classification accuracy (F1=0.23), while adding accelerometer features improved performance to 78.7% accuracy (F1=0.74) - a 42.6 percentage point gain - confirming that accelerometer-assisted artifact labeling is essential for reliable signal quality assessment. These results do not validate the nail-mounted device directly; rather, they demonstrate the feasibility of the proposed artifact-detection pipeline on publicly available PPG data. This framework provides an early structured, technology-grounded proposal for wearable vital-sign monitoring in the Azerbaijani post-conflict resettlement context.