Machine Learning Based IoT Adaptive Architecture for Epilepsy Seizure Detection: Anatomy and Analysis

TL;DR

This paper presents an affordable, noninvasive, real-time machine learning system using k-Nearest-Neighbors for personalized epilepsy seizure detection, achieving high accuracy with quick adaptation for individual users.

Contribution

It introduces a simple, adaptable, and fast-training seizure detection system based on kNN, validated on a large dataset of 500 subjects.

Findings

Mean accuracy of 94.5% in seizure detection

Training time less than four seconds per user

Validated on 500 subjects with 178 Hz sampling

Abstract

A seizure tracking system is crucial for monitoring and evaluating epilepsy treatments. Caretaker seizure diaries are used in epilepsy care today, but clinical seizure monitoring may miss seizures. Monitoring devices that can be worn may be better tolerated and more suitable for long-term ambulatory use. Many techniques and methods are proposed for seizure detection; However, simplicity and affordability are key concepts for daily use while preserving the accuracy of the detection. In this study, we propose a versal, affordable noninvasive based on a simple real-time k-Nearest-Neighbors (kNN) machine learning that can be customized and adapted to individual users in less than four seconds of training time; the system was verified and validated using 500 subjects, with seizure detection data sampled at 178 Hz, the operated with a mean accuracy of (94.5%).