A logistic regression-based risk screening framework for early malnutrition detection in children under five

Malnutrition among children under five years of age remains a leading cause of preventable morbidity and mortality in Nigeria and across low- and middle-income countries. Existing detection methods are resource-intensive, dependent on trained clinical personnel, and largely inaccessible in the underserved communities where malnutrition burden is highest. This paper presents the design, implementation, and evaluation of a cloud-deployed, supervised machine learning system for early malnutrition risk screening. A Logistic Regression model was trained on 50,000 synthetic pediatric samples generated using WHO West African growth reference values, incorporating 12 features encompassing anthropometric measurements, WHO-derived z-scores (HAZ, WAZ, WHZ), clinical status flags, and socioeconomic indicators. The system was evaluated using 5-fold stratified cross-validation and achieved an Area Under the Receiver Operating Characteristic Curve (AUC) of 0.7035 ± 0.0256, demonstrating acceptable discriminative ability for a community-level prototype screening tool. The deployed web application delivers single-patient risk classifications (HIGH, MODERATE, or LOW) in under 500ms and supports batch processing of up to 10,000 records via CSV upload. The system’s architecture — Flask 3.x backend on Render, React 18 frontend on Netlify — operates within free-tier cloud infrastructure, making it viable for deployment without capital expenditure. Results demonstrate the system’s acceptability as an interpretable, low-cost decision-support tool for community health workers operating in resource-limited settings.