Application of machine learning algorithms to predict early childhood development in children aged 24-59 months across three East African countries.

backgroundEarly childhood development (ECD) plays a crucial role in shaping the future development of children and it influences their lifelong outcomes. The Early childhood development index 2030 (ECDI2030) serves as an effective tool for monitoring the overall development of children aged 24-59 months at the population level. This study employed machine learning algorithms to identify the predictors of ECD across three East African countries, using the ECDI2030.

methodsData were derived from the Demographic and Health Surveys of Kenya, Mozambique, and Tanzania. Seven supervised machine learning algorithms and an ensemble of the best performing models were utilized to predict ECD. The dataset was randomly divided into 80% training and 20% testing sets. The predictive ability of each machine learning model was evaluated using area under the curve (AUC) and the classification metrics. We used SHapley Additive exPlanations (SHAP) to explain the predictions by interpreting feature importance.

resultsAbout 57.4% (95% CI = 56.5, 58.3) children were developmentally on track in health, learning, and psychosocial well-being. The ensemble model of extreme gradient boosting and random forest was the best algorithm with accuracy of 66% and AUC of 71%. The top three most important predictors of ECD were child age, media exposure, and maternal education level with a mean absolute SHAP value of +0.17, + 0.12, and +0.1, respectively. The beeswarm plot of SHAP revealed that children aged 24-35 months, those whose mothers were not exposed to media, or those whose mothers had completed at least secondary education were more likely to be developmentally on track.

conclusionIn East Africa, only the modest majority of children were developmentally on track. Policies should prioritize preprimary education, equitable access, and women's education to empower mothers and improve parenting practices. Promoting appropriate media use while limiting maternal screen time can enhance children's developmental outcomes in East Africa and other countries with similar socioeconomic contexts, including most sub-Saharan African countries.