What goes on when the lights go off? Using machine learning techniques to characterize a child's settling down period

Area of Science:

• Pediatric Sleep Research
• Machine Learning in Healthcare
• Sensory Processing in Children

Background:

• Objective sleep measurement in children often neglects the crucial pre-sleep 'settling down' period.
• Sensory sensitivities, particularly to tactile input, can influence a child's pre-sleep behavior.
• Existing metrics may not adequately capture subtle activity differences during this transitional phase.

Purpose of the Study:

• To investigate pre-sleep activity patterns in children with and without sensory sensitivities (SS).
• To apply machine learning to identify objective features differentiating these groups during the settling down period.
• To explore the potential of settling down activity as a novel target for understanding sleep disturbances in children with SS.

Main Methods:

• Actigraphy data collected over two weeks from 17 children with SS and 18 without (NSS).
• Analysis focused on the settling down period, extracting seven distinct activity features.
• 10-fold cross-validation with random forests used for group differentiation accuracy assessment.

Main Results:

• Machine learning accurately differentiated groups with 83% accuracy, 83% specificity, and 84% sensitivity.
• Children with SS exhibited significantly higher maximum activity bouts and greater activity variance (e.g., interquartile range, Shannon entropy) during settling down.
• These specific activity features distinguished children with SS from their peers.

Conclusions:

• Novel machine learning approach successfully identified differentiating activity features during the settling down period.
• These pre-sleep activity differences are not easily captured by standard sleep metrics.
• Activity during the settling down phase represents a promising area for future research into sensory sensitivities and sleep in children.