Predicting nonresponse to sexual identity question in youth risk behavior surveillance: A machine learning analysis of complex survey data

PURPOSE: To compare seven machine learning (ML) models developed to predict non-response to the sexual identity question in the 2023 Youth Risk Behavior Surveillance System (YRBSS) and identify the best-performing ML model, along with key attributes associated with the non-response. METHODS: Data of 20,103 students, with 32 predictors across domains of personal characteristics, school behavior, substance use, and sexual activity were analyzed. Supervised ML models-including random forest (RF), gradient boosting, extreme gradient boosting, decision tree, neural network, lasso, and elastic net were developed and incorporated survey weights. Performance was assessed using F1 score, area under the ROC curve (AUC), and area under the precision-recall curve (AUPRC). RESULTS: About 10 % of students didn't respond to the sexual identity question, with higher rates among racial/ethnic minorities, including American Indian/Alaska Native and Native Hawaiian/Pacific Islander youths. RF model showed the most robust overall performance across all metrics. Attributes predicting non-response included response status to questions of school absence due to safety concerns and having ≥ 4 sexual partners. CONCLUSIONS: Non-response was non-random and concentrated among vulnerable groups. Predictive performance was strong, but findings suggest that response patterns to other sensitive survey items play substantial role, with implications for survey design and non-response adjustment.