Neural Network-Based Prediction of Perceived Sleep Quality Through Wearable Device Data

Area of Science:

• Biomedical Engineering
• Computational Neuroscience
• Sleep Science

Background:

• Investigated the use of wearable devices to collect physiological data for sleep quality assessment.
• Collected heart rate, physical activity, and sleep quality data (device-measured and self-reported) from 18 participants over four weeks.

Purpose of the Study:

• To develop and evaluate a neural network model for predicting perceived sleep quality using wearable sensor data.
• To establish correlations between objective physiological metrics and subjective sleep quality perceptions.

Main Methods:

• Employed data processing and feature engineering techniques.
• Optimized a Multi-Layer Perceptron (MLP) classifier for sleep quality prediction.
• Utilized physiological data (heart rate, activity) and self-reported sleep quality.

Main Results:

• The neural network model achieved a moderate predictive accuracy of 59% for perceived sleep quality.
• When a 1-grade tolerance (on a 1-5 scale) was applied, the accuracy increased to 92%.
• Results indicate challenges in precisely quantifying subjective sleep experiences from wearable data alone.

Conclusions:

• Further research is needed to fully understand the role of wearables and artificial intelligence in sleep behavior analysis.
• The findings suggest that while direct prediction is challenging, AI models with tolerance thresholds show promise.
• Wearable technology and AI may offer valuable tools for augmenting sleep research and understanding individual sleep patterns.