A recent cross-sectional study conducted in Heilongjiang, a province in northeastern China, has used machine learning techniques to identify preschoolers with insufficient sleep. The research, led by Xichao Zhang of Harbin Sport University and published in BMC Public Health, analyzed data from 1,258 children aged three to six years collected in 2020. Researchers utilized routinely collected data, including physical activity levels, screen time, body size, and fitness test results, rather than relying solely on sleep diaries or actigraphy.
Read More
The study found that 344 children, representing 27.3 percent of the sample, did not meet recommended sleep duration guidelines. This highlights a concerning trend, as more than one in four preschoolers in the study area are not getting adequate sleep, aligning with global worries regarding children’s sleep health.
To ensure methodological rigor, the researchers excluded variables with more than 15 percent missing data and divided participants into training and validation sets, maintaining similar proportions of short and adequate sleepers in both groups. Advanced techniques were applied to handle missing values and select features, resulting in six key predictors of insufficient sleep, including weekend physical activities and body mass index.
The study employed seven classifiers to identify sleep insufficiency, finding that the Light Gradient Boosting Machine achieved the highest predictive accuracy. However, the random forest model was favored for its balanced performance, successfully distinguishing well-rested children from those with sleep deficits, albeit with reduced sensitivity.
Using Shapley additive explanations (SHAP), the study also revealed nonlinear relationships between various factors and sleep outcomes, suggesting that different patterns of physical activity are associated with varying risks of sleep insufficiency. These insights could guide targeted interventions that promote outdoor activity and play on weekends, which might contribute to improved sleep among preschoolers.
Importantly, the authors emphasize that their findings serve as a tool for further research rather than a substitute for direct sleep assessments. The caregiver-reported nature of sleep data introduces potential biases, and they stress the need for validation across different regions and contexts before broad application.
This research highlights the complex interplay between sleep, physical activity, and overall child health, presenting an innovative approach that could be adapted to monitor and improve sleep health in similar populations.