Machine Learning Approach for Real-Time Fatigue Monitoring Using IMU Data and Mobile Application

Abstract

Accurately monitoring fatigue in neurodegenerative conditions such as Ataxia-Telangiectasia (A-T) remains a challenge when using traditional clinical methods. This project investigated a real-time monitoring system that combines wearable IMU sensors with a mobile application for subjective self-reporting of fatigue, based on Ecological Momentary Assessment (EMA). The aim was to explore whether integrating movement-based features with real-time fatigue input could provide a more comprehensive and responsive understanding of fatigue in daily life. Data were collected from three healthy adults over seven days, using ankle- and waist-mounted sensors and an emoji-based mobile app. Features such as jerk, entropy, and variability were extracted from the IMU data, and Random Forest classifiers were used to classify fatigue levels. Findings showed that individual differences strongly influenced model performance, and that combining movement data with self-reports provided a more comprehensive view of fatigue. Several technical and practical challenges emerged, including data loss, app limitations, and inconsistent participant logging. These issues demonstrated the need for more flexible and robust systems that can adapt to each user’s behaviour and environment. The results support the feasibility of personalised fatigue monitoring in natural settings and offer a foundation for future research with clinical populations. Further development should involve children with A-T, extend the monitoring period, and explore adaptive or contactless sensing approaches to improve long-term usability.