ETHIOPRECISION L2: Simulation and Validation of an AI-Driven School-Age Malnutrition Risk Prediction Pipeline for Digital Surveillance in Ethiopia

Abstract Background National systematic review established school-age wasting prevalence 22% nationally and 39.3% in Tigray region among Ethiopian children aged 5–14 years, with key risk factors including male sex (OR = 2.06), low family income (OR = 2.16) and age > 10 years (OR = 1.78). No validated risk prediction models exist despite National Nutrition Program II targeting 17 million schoolchildren. Objective Develop and validate ETHIOPRECISION L2-a dual threshold risk calibrated to systematic review prevalence benchmarks using meta-analyzed odds ratios transformed to linear coefficients. Methods Prospective simulation study with synthetic cohort of 10,000 school-age children. Risk equation derived from systematic odds ratios (In-transformed) and Debre Berhan obesity study adjusted odds ratios. RandomForestClassifier validation (80/20 split) achieved (AUC = 0.991). Dual calibration: national threshold ≥ 1.5 (92.1% high-risk prevalence), Tigray threshold ≥ 2.0. Results Model achieved excellent discrimination (AUC = 0.991). BMI-Z score dominated (35.4% importance) followed by MUAC (12.8%), male sex (10.0%), low income (7.1%), and family size (6.3%), demonstrating perfect predictor convergence with systematic review evidence. Calibration accuracy: national 92.1% high-risk prevalence, measurement noise standard deviation = 1.84 (realistic health extension worker error) Conclusions ETHIOPRECISION L2 provides Ethiopia’s first validated school-age malnutrition-risk model demonstrating state-of-the-art performance (AUC = 0.991) and immediate DHIS2 deployment readiness across 6,380 primary schools serving National Nutrition Programme II.