Affordable ECG Sensing: A Machine Learning-Enhanced Framework for Scalable Human-Machine Interaction in Telehealth

Affordable and accurate cardiovascular monitoring is a pervasive issue in resource-constrained and rural settings, where sophisticated clinical-grade electrocardiogram (ECG) devices are exorbitantly expensive. Low-cost ECG sensors available today are prone to motion artifacts and noisy signals, limiting their use to heart rate variability (HRV) estimation. To fill this gap, we present a proof-of-concept system calibrating noisy, wearable ECG-based HRV measurements using machine learning (ML) models, against a gold standard clinical-grade sensor. Our objective is to evaluate the feasibility of accurate HRV prediction i.e., SDNN and RMSSD using low-cost sensors (AD8232, MAX30003) using ensemble and regression-based ML models. Data were collected in semi-controlled wearable settings, digital-filtered preprocessed, and processed with model ensembles optimized for generalizability. The top-performing models yielded RMSEs of 14.09 ms (SDNN) and 9.09 ms (RMSSD), with sensor type variable performance. In addition to signal calibration, the system includes a real-time human–machine interface (HMI) with wireless ECG visualization and feedback, integrated by an IoT-capable microcontroller. While preliminary and unclinically validated, our results suggest ML-augmented low-cost ECG systems to be feasible for scaled telehealth and affect-aware applications.