Predicting undernutrition among elementary schoolchildren in the Philippines using machine learning algorithms

Area of Science:

• Nutritional epidemiology
• Machine learning applications in public health
• Dietary assessment

Background:

• Undernutrition remains a significant public health challenge, particularly in developing countries.
• Accurate prediction of undernutrition is crucial for effective intervention strategies.
• Existing methods for classifying nutritional risk may not fully capture the complexities of dietary intake.

Purpose of the Study:

• To compare the accuracy of four machine learning algorithms in predicting undernutrition among Filipino schoolchildren.
• To evaluate two classification schemes, National Academy of Medicine (NAM) acceptable macronutrient distribution ranges (AMDRs) and Philippine Dietary Reference Intakes (PDRIs), for nutritional risk assessment.
• To assess the performance of machine learning predictions against anthropometric classifications used by the national school feeding program.

Main Methods:

• Data from 618 public-school children in the Philippines were collected using 24-h dietary recalls and socioeconomic surveys.
• Nutritional risk was classified using NAM AMDRs and PDRIs.
• Four machine learning algorithms (random forest, support-vector machine, linear discriminant analysis, logistic regression) were employed to predict undernutrition.
• Algorithm performance was evaluated based on accuracy, sensitivity, and specificity.

Main Results:

• The prevalence of undernutrition was higher under NAM AMDRs (82.67%) than PDRIs (78.71%).
• The random forest algorithm demonstrated the highest accuracy (78.55%) in predicting undernutrition.
• Key predictors identified by random forest included household expenditures, child/household age, food insecurity, and dietary diversity.
• AMDR classifications identified more children at risk (477) compared to anthropometric classifications (213).

Conclusions:

• The random forest algorithm is a promising tool for predicting undernutrition in Filipino schoolchildren.
• AMDR classification can aid in targeting beneficiaries for feeding programs.
• Incorporating local dietary culture and addressing data representation for marginalized populations are essential for effective nutrition interventions.