Rural-Urban Disparities in Early Childhood Development: Insights from Interpretable Ensemble Machine Learning Model

Abstract Background Early childhood development (ECD) is essential for lifelong health, education, and economic productivity, as emphasized by SDG Target 4.2, which seeks universal access to quality pre-primary education by 2030. In Nigeria, ECD are significantly impacted by poverty, malnutrition, and systemic inequities. However, there is limited research on context-specific determinants such as caregiving practices, parental discipline, and home environments. This study developed a Multilayer Stacked Weighted Ensemble Model (MSWEM) and analyse data from the 2021 Nigeria Multiple Indicator Cluster Survey (MICS), comprising 15,899 children aged 24–59 months. Statistical techniques included descriptive analysis, binary logistic regression for predictor identification and machine learning for advanced prediction and SHAP (SHapley Additive exPlanations) analysis for model interpretability. The model development was implemented in Python 3.7, integrating ensemble methods to enhance prediction accuracy Results The MSWEM achieved a high predictive accuracy with an ROC-AUC of 93% and accuracy of 85.2%. The model correctly predicted 46.7% on – track children in urban areas and 24.7% on-track children in rural areas. Factors influencing the likelihood of a child being on track include age of the child, cognitive caregiving practice, overall physical home environment, mother’s education, age of woman, to mention but a few Conclusion This study highlights the utility of interpretable machine learning in understanding and predicting ECD outcomes. These efforts are essential for ensuring equitable opportunities for all children in Nigeria, aligning with national development goals and global progress toward SDG 4.2.