Portable IoT Device with Embedded Machine Learning for Real-Time Heart-Rate and SpO_2 Vital-Sign Monitoring and Classification

Pacheco-Anton F.D.L.A.; Pauca-Muñoz F.A.; Borromeo-Quispe A.J.; Lozano-Villalobos A.J.; Pariona-Sanchez J.M.; Cabana-Cáceres M.R.; Castro-Vargas C. · Instrumentation Mesure Metrologie (2026) · DOI: 10.18280/i2m.250402

Resumen

This paper presents the design, implementation, and preliminary validation of a low-cost portable device for real-time monitoring of basic vital signs, namely heart rate and peripheral oxygen saturation (SpO₂), intended for preventive cardiovascular screening outside specialized clinical settings. The system integrates an ESP32 microcontroller with a MAX30102 photoplethysmography (PPG) sensor and a TFT display, and embeds a TensorFlow Lite Micro classifier mapping four features into five descriptive categories, with values transmitted over Wi-Fi to a cloud database. Bench accuracy was assessed against a commercial Food and Drug Administration (FDA)-cleared pulse oximeter (Contec CMS50D). In a short single-session comparison, the prototype showed a mean absolute percentage error (MAPE) of 1.8% for SpO₂ and 2.3% for heart rate, sub-500 ms latency, and six hours of autonomy. These results are derived from a single-session, single-participant comparison and are preliminary and not yet confirmed on a larger sample. The embedded classifier executed correctly in real time but is presented as a functional demonstrative module, since formal accuracy metrics on an independent test set were not collected. The work integrates real-time PPG acquisition, cloud telemetry, and on-device classification on a single low-cost platform, separating preliminary accuracy figures from the classifier's unvalidated status. Results suggest the architecture is a technically promising, economically accessible alternative for continuous monitoring in resource-constrained settings, pending larger-scale validation.

Instituciones: Universidad Nacional Federico Villarreal