Machine learning-based insomnia symptom risk classification model for residents of Hebei province.

OBJECTIVES: The aim is to develop and externally validate a machine learning-based risk classification model for insomnia symptoms in a community-based population sample, and to identify and explain the main contributing factors using shapley additive explanations (SHAP). STUDY DESIGN: Cross-sectional surveys were conducted in 2021 and 2024 across 30 randomly selected counties within Hebei Province, China, enrolling 16,848 eligible participants. METHODS: Insomnia symptoms were assessed using the Insomnia Severity Index (ISI). The 2021 dataset was randomly split (7:3) into a training set (n = 6863) and an internal validation set (n = 2942). Six algorithms were used to construct risk classification models. The optimal model was subsequently selected and interpreted using SHAP. RESULTS: Among the six models evaluated, the logistic regression model demonstrated superior classification performance, achieving AUC values of 0.812 in the internal validation set and 0.795 in the external validation set. Furthermore, the SHAP analysis revealed that the top five influential features in the model were sleep time (SHAP value: 0.071), sleep latency (0.067), bedtime (0.061), health status (0.044), and age (0.026). CONCLUSION: The insomnia symptom risk classification model developed with the logistic regression algorithm demonstrated satisfactory performance. SHAP improves the interpretability of the model.