Uncovering the determinants of infant and toddler childcare demand in less-developed rural China: a machine learning perspective

Introduction: China faces a severe imbalance between the supply and demand of formal childcare services for infants and toddlers aged 0-3, with rural research and resource allocation falling far behind urban areas. Objective: This study aimed to identify key determinants of rural childcare demand in western China's less-developed regions and to screen the optimal predictive machine learning (ML) prediction model, based on Andersen's Behavioral Model. Methods: A cross-sectional survey with purposive and multi-stage sampling was conducted in southwestern Guizhou Province, collecting valid data from 1,116 rural families with infants and toddlers. Seven ML algorithms were applied to construct childcare demand prediction models with comprehensive model evaluation, SHAP analysis for feature identification, and threshold analysis for key factors performed subsequently. Results: The random forest (RF) model was identified as the optimal model, demonstrating robust generalization and discriminative ability. SHAP analysis revealed childcare flexibility, overall childcare quality, early education, and teacher professionalism as the four core positive determinants of rural childcare demand. Threshold analysis further defined the optimal critical values of these key factors and verified their strong practical predictive performance, with childcare flexibility and overall quality showing the most prominent effects. Conclusion: This study confirms that machine learning can effectively identify the determinants of rural childcare demand. The four service-related factors are the most influential drivers. The findings provide empirical evidence for optimizing rural childcare services and formulating demand-oriented childcare policies to promote the high-quality and inclusive development of rural childcare systems in less-developed regions in China.