TinyML-Based IoT Wearable for Seizure-Like Event Detection and Emergency Alerting

Abstract

Epilepsy affects a large population worldwide, and a considerable share of seizure-related injuries occur when an episode is unwitnessed. This paper presents the proposed architecture of a low-cost, wrist/limb-worn wearable device intended to flag seizure-like physical movement and alert a designated caregiver. The system combines a six-axis inertial measurement unit (IMU) for motion sensing with a photoplethysmography (PPG) sensor as a supplementary physiological input, processed on an ESP32 microcontroller. Two detection paths are proposed: a TinyML classifier trained on motion-pattern features, and a deterministic threshold-based fallback rule that can operate independently if the learned model is unavailable or under-trained at the time of deployment. On flagging a probable event, the device is designed to acquire a GPS location fix and transmit an SMS alert through a GSM module to a caregiver. The device does not diagnose epilepsy and does not claim clinical-grade seizure detection; it is designed strictly as a supportive alerting aid that flags seizure-like movement patterns for human verification. This paper describes the system architecture, sensor roles, decision logic, and the validation plan intended for the working-prototype stage of the project.