Forecasting under-five stunting in Ethiopia using classical and machine learning time series models

Abstract

Background The prevalence of under-five stunting in Ethiopia remains above 30%, which, according to World Health Organization (WHO), constitutes a major public health concern. Stunting has long-term consequences for child growth, development, and overall health. Accurate forecasting of its prevalence is therefore essential to guide policymakers and inform targeted interventions. This study aimed to forecast the prevalence of under-five stunting in Ethiopia for the period 2025–2030 using historical data and time series modeling. Methods Annual under-five stunting prevalence data for Ethiopia from 2000 to 2024 were retrieved from the WHO Global Health Observatory. Time series forecasting models, including Autoregressive Integrated Moving Average (ARIMA), Exponential Smoothing (ETS), Multilayer Perceptron (MLP), and Long Short-Term Memory (LSTM), were developed and evaluated. Model performance was assessed using time series cross-validation with mean absolute error (MAE), mean absolute percentage error (MAPE), and R² as evaluation metrics. The best-performing model was applied to forecast stunting prevalence for 2025–2030. Results The ETS model demonstrated the best predictive performance (MAE = 1.09, MAPE = 2.71%, R² = 0.903) and was selected for forecasting. Forecasting indicated a gradual decline in under-five stunting prevalence in Ethiopia from 33.95% in 2025 to 31.95% in 2030. The projected prevalence for 2029 is 32.35% (95% CI: 26.90–37.79%), above the national target of 19%, and the 2030 forecast remains well above the SDG target of ending all forms of malnutrition. Conclusion and recommendation Under-five stunting in Ethiopia is projected to remain above national and SDG targets by 2030, indicating current nutrition efforts are insufficient; national program evaluation and evidence-based policy adjustments are recommended.