Benson FN
Odhiambo R
Ngugi AK
Brink W
Waljee AK
Moyer CA
Zhu J
Agoi F
Abubakar A
Pediatr Res. 2026;
BACKGROUND: This study aimed to address the critical gap in the limited application of machine learning (ML) for identifying developmental delays in low-resource settings by developing models to predict off-track development in infants aged 0 to 6 months and identify key predictors. METHODS: A cross-sectional study involving 1,995 singleton infants aged 0 to 6 months was conducted in Kaloleni and Rabai sub-counties, Kilifi, Kenya, between March 2023 and March 2024. Development was assessed using the World Health Organization's Indicators of Infant and Young Child Development tool, with Development-for-Age Z-scores used to classify infants as on- or off-track. Ridge logistic regression (LR), random forest (RF), and extreme gradient boosting (XGBoost) models were trained using sociodemographic, psychosocial, clinical/biological, nutritional, and health-related predictors. Performance was evaluated using area under the receiver operating characteristic curve (AUC), accuracy, sensitivity, and specificity. SHapley Additive exPlanations enhanced model interpretability. RESULTS: Approximately 10.4% of infants were developmentally off-track. Ridge LR, RF, and XGBoost showed similar performance, with AUCs of 76.6%, 75.8%, and 76.1%, respectively. Limited psychosocial stimulation and increasing infant age were the strongest predictors. CONCLUSIONS: This study highlights the burden of developmental delays in low-resource settings. ML models show promise for early risk prediction and targeted intervention, though further validation is recommended. IMPACT: Early intervention programs are proven to enhance optimal childhood development, yet the vital step of early identification of developmental delays is often overlooked. While machine learning is increasingly used to predict or identify health outcomes, its application in identifying developmental outcomes, particularly in low-resource settings, remains limited. This study contributes to the literature by applying machine learning to identify infants who are developmentally off-track and highlights key predictors. Limited psychosocial stimulation and increasing infant age were the strongest predictors, alongside low socioeconomic status, maternal mental health challenges, limited healthcare access, and nutritional and biological risks.
