Wearable-oriented Support for Interpretation of Behavioural Effects on Sleep

Area of Science:

• Health Sciences
• Computer Science
• Biotechnology

Background:

• Daily behaviors significantly influence short-term and long-term health outcomes.
• Adopting healthy habits is crucial for preventing or delaying chronic diseases.
• Understanding current behavioral patterns is the first step toward positive change.

Purpose of the Study:

• To develop a strategy for interpreting longitudinal wearable data to analyze health issue causes.
• To use the sleep domain as a case study to identify factors contributing to poor sleep quality.
• To create an explainable deep learning (DL) model for analyzing wearable data.

Main Methods:

• Utilized a dataset comprising 1874 days of wearable data.
• Developed an explainable deep learning (DL) model to identify day-before-night behaviors affecting sleep.
• Validated findings through comparative analysis with a hormone-based sleep control framework.

Main Results:

• The explainable DL model successfully identified key behaviors preceding poor sleep quality.
• The model's explanations aligned with existing literature and hormone-based sleep control frameworks.
• Identified specific day-before-night behaviors as potential causes of diminished sleep quality.

Conclusions:

• Interpreting multifeatured longitudinal wearable data with explainable DL can reveal causes of health issues like poor sleep.
• Findings support the link between specific daily behaviors and sleep quality, corroborating existing research.
• Further research with more comprehensive datasets is recommended to explore feature combinations and their health impacts.