Skip to content
Open access

A User Fatigue Prediction System Based on Heart Rate and SpO₂ Data from Smartwatch on Android Platform

Oct 2026 · Journal of Artificial Intelligence and Engineering Applications (JAIEA) · 0 citations · 19 references

Abstract

The high demands of modern activities frequently lead to fatigue, a condition that diminishes physical and cognitive capacity while elevating the risk of human error. Smartwatches equipped with physiological sensors offer continuous, non-invasive health monitoring, presenting an opportunity for early fatigue detection. This study developed an Android application that predicts user fatigue status by analyzing heart rate (BPM) and blood oxygen saturation (SpO₂) data recorded by a smartwatch. A machine learning model based on Long Short-Term Memory (LSTM) was trained on a public dataset to recognize fatigue patterns, then converted to TensorFlow Lite format for on-device integration. The application retrieves historical BPM and SpO₂ data through Health Connect API and classifies the user's condition as either Normal or Fatigued. Offline evaluation yielded 100% accuracy on test data; however, this performance was identified as an artifact of overfitting caused by deterministic synthetic labeling rules. On-device inference encountered compatibility issues with TensorFlow Lite runtime, preventing successful execution of the LSTM model on Android. Despite this limitation, the study demonstrates the feasibility of bridging heterogeneous wearable ecosystems and highlights critical technical challenges in deploying recurrent neural networks on mobile devices.

Read PDF

We use cookies to run the site and, with your consent, for analytics and to show ads. See our Cookie Policy.