A Wearable IoT Device with Embedded Risk Prediction for Febrile Seizures in Children

Abstract

A febrile seizure (FS) occurs in 2–5% of children aged 6 months to 5 years, and up to 40% will have a recurrence. The design, implementation, and initial validation of a cost-effective, wearable, IoT-based system for real-time FS risk prediction are presented. The proposed system integrates two physiological sensors (B3950 NTC thermistor for body temperature; MAX30102 for heart rate) with caregiver-entered clinical and demographic data (age, gender, family history of FS, maternal smoking history), all processed by an ESP32 microcontroller. An embedded logistic regression model classifies risk into three levels (Low / Moderate / High) and generates Wi-Fi alerts via a web interface. The model was trained on a 5,000-case synthetic dataset developed from published FS epidemiological literature using Monte Carlo simulation. On a stratified locked synthetic test set (n = 1,500), the model achieved an AUC of 0.9438, sensitivity of 85.00%, specificity of 87.46%, NPV of 99.04%, and an F1-score of 41.72% at the Youden-optimal threshold of 0.0740. Feature analysis confirmed body temperature and family history as the strongest predictors, consistent with published clinical evidence. End-to-end prototype testing verified correct subsystem operation. A practical comparison of the wearable temperature subsystem against a certified digital clinical thermometer yielded a mean error of −0.35 °C, supporting the measurement consistency of the sensing subsystem. The results show the engineering feasibility of an inexpensive, interpretable, wearable proof-of-concept prototype for monitoring FS. Before deployment, clinical applicability in home or resource limited settings needs prospective validation on real pediatric clinical data.