Predicting off-track development in infants aged 0-6 months in low-resource settings using machine learning.

Area of Science:

• Pediatric Health
• Machine Learning Applications
• Global Child Development

Background:

• Developmental delays affect many children, particularly in low-resource settings.
• Early identification and intervention are crucial for optimal childhood development.
• Machine learning (ML) applications for identifying developmental delays are limited in low-resource contexts.

Purpose of the Study:

• To develop and evaluate ML models for predicting developmental delays in infants aged 0-6 months.
• To identify key predictors of off-track development in infants within low-resource settings.
• To address the gap in ML application for early childhood developmental assessment.

Main Methods:

• A cross-sectional study of 1,995 infants (0-6 months) in Kenya.
• Development assessed using WHO Indicators of Infant and Young Child Development tool.
• Ridge logistic regression, random forest, and extreme gradient boosting models were trained and evaluated.

Main Results:

• 10.4% of infants were identified as developmentally off-track.
• ML models (Ridge LR, RF, XGBoost) demonstrated comparable performance (AUCs ~76%).
• Key predictors included limited psychosocial stimulation and increasing infant age.

Conclusions:

• ML models show promise for early risk prediction of developmental delays in low-resource settings.
• Findings highlight the burden of developmental delays and the need for targeted interventions.
• Further validation of ML models for developmental assessment is recommended.