Machine learning characterizes insufficient sleep among preschool children in northeastern China

A cross-sectional study in BMC Public Health examined whether routine behavioral, body-size, and fitness measures could characterize caregiver-reported insufficient sleep among preschool children in cold-region China. Researchers analyzed 2020 monitoring and caregiver-questionnaire data from 1,258 children aged 3–6 years in Heilongjiang. They used an outcome-stratified 7:3 training–validation split and evaluated seven classifiers, restricting preprocessing, feature selection, and tuning to training data.

Insufficient sleep was reported for 344 children (27.3%). Six retained predictors comprised BMI, weekend moderate-to-vigorous activity, weekend outdoor activity, weekend screen time, standing long jump, and two-leg consecutive jumping time. Light Gradient Boosting Machine achieved the highest area under the receiver operating characteristic curve (AUC), at 0.849 (95% CI, 0.806–0.889). The random forest model selected for interpretation had an AUC of 0.827 (95% CI, 0.781–0.872), versus 0.752 (95% CI, 0.692–0.806) for logistic regression. Random forest sensitivity was 55.3% and specificity 89.1%.

Model explanations ranked weekend outdoor activity, weekend moderate-to-vigorous activity, and BMI highest, with nonlinear relationships to model output. These findings support hypothesis generation rather than causal conclusions or replacement of direct sleep assessment. Caregiver-reported outcomes, cross-sectional data, and internal-only validation limit clinical application; validation across regions, climates, and seasons remains necessary.