A Positionally Encoded Transformer for Monitoring Health Contexts of Hajj Pilgrims from Wearable Sensor Data

Area of Science:

• Wearable sensor technology
• Artificial intelligence in healthcare
• Human activity recognition

Background:

• Real-time health monitoring is crucial during physically demanding activities like the Hajj pilgrimage.
• Existing methods struggle to accurately detect diverse health contexts such as physical fatigue, emotional state, and activity type.

Purpose of the Study:

• To develop and evaluate a novel positionally encoded Transformer model for real-time detection of health-relevant contexts from wearable sensor data.
• To improve the accuracy and robustness of health monitoring systems for individuals undertaking strenuous physical tasks.

Main Methods:

• Utilized a positionally encoded Transformer model incorporating Long Short-Term Memory (LSTM) for feature extraction.
• Employed Transformer layers for context classification, leveraging positional encoding to analyze sequential sensor data.
• Conducted experiments with data from 19 participants during physically demanding tasks.

Main Results:

• The proposed model demonstrated high classification accuracy across multiple health-relevant contexts, including physical tiredness, emotional mood, and activity type.
• Significant improvements in real-time health monitoring capabilities were observed compared to baseline methods.
• The model effectively captured sequential dependencies within time-series sensor data.

Conclusions:

• The positionally encoded Transformer model offers a robust and accurate solution for real-time health monitoring using wearable sensors.
• This approach enhances the ability to understand and respond to individual health states during physically demanding activities.
• The findings have significant implications for improving participant safety and well-being in challenging environments.