Model and variable selection using machine learning methods with applications to childhood stunting in Bangladesh

Area of Science:

• Public Health
• Data Science
• Pediatrics

Background:

• Childhood stunting remains a significant public health issue in Bangladesh.
• Conventional statistical methods have limitations in analyzing complex health data for stunting risk factors.
• Machine learning (ML) applications for predicting stunting are underexplored.

Purpose of the Study:

• To evaluate the performance of various ML methods in predicting childhood stunting.
• To identify key variables contributing to stunting in under-5 children in Bangladesh.
• To assess the utility of ML in understanding stunting determinants.

Main Methods:

• Utilized data from the 2014 Bangladesh Demographic and Health Survey.
• Applied and compared several ML algorithms: gradient boosting, random forests, support vector machines, classification tree, and logistic regression.
• Identified important predictive variables for childhood stunting.

Main Results:

• Gradient boosting demonstrated the lowest misclassification error in predicting stunting.
• The top 10 predictive variables include child age, wealth index, maternal education, preceding birth interval, paternal education, division, household size, maternal age at first birth, maternal nutritional status, and parental age.
• ML models successfully identified significant demographic, socioeconomic, nutritional, and environmental factors.

Conclusions:

• Machine learning methods are valuable tools for building accurate prediction models for childhood stunting.
• Demographic, socioeconomic, nutritional, and environmental factors are crucial for understanding and addressing stunting in Bangladesh.
• The findings support targeted interventions based on identified risk factors.