Additional file 1 of Can we predict the burden of acute malnutrition in crisis-affected countries? Findings from Somalia and South Sudan

Additional file 1: Figure S5. Causal framework for acute malnutrition among children, used to identify potential predictors. Figure S6. GLM-predicted versus observed SAM (MUAC + oedema) prevalence, Somalia, by district-month, on training data, LOOCV and holdout data. Shaded channels indicate different absolute deviance of predictions. Vertical dotted lines denote commonly used SAM prevalence thresholds. Figure S7. GLM-predicted versus observed GAM (WFH + oedema) prevalence, Somalia, by district-month, on training data, LOOCV and holdout data. Shaded channels indicate different absolute deviance of predictions. Vertical dotted lines denote commonly used GAM prevalence thresholds. Figure S8. GLM-predicted versus observed GAM (MUAC + oedema) prevalence, Somalia, by district-month, on training data, LOOCV and holdout data. Shaded channels indicate different absolute deviance of predictions. Vertical dotted lines denote commonly used GAM prevalence thresholds. Figure S9. GLM-predicted versus observed mean WFH, Somalia, by district-month, on training data, LOOCV and holdout data. Shaded channels indicate different absolute deviance of predictions. Vertical dotted lines denote potentially useful thresholds. Figure S10. GLM-predicted versus observed mean MUAC, Somalia, by district-month, on training data, LOOCV and holdout data. Shaded channels indicate different absolute deviance of predictions. Vertical dotted lines denote potentially useful thresholds. Table S7. Performance of random forest models in Somalia, by acute malnutrition outcome. Figure S11. RF-predicted versus observed GAM (WFH + oedema) prevalence, Somalia, by district-month, on training data, LOOCV and holdout data. Shaded channels indicate different absolute deviance of predictions. Vertical dotted lines denote commonly used GAM prevalence thresholds. Figure S12. RF-predicted versus observed mean WFH, Somalia, by district-month, on training data, LOOCV and holdout data. Shaded channels indicate different absolute deviance of predictions. Vertical dotted lines denote potentially useful thresholds. Figure S13. GLM-predicted versus observed SAM (MUAC + oedema) prevalence, South Sudan, by district-month, on training data, LOOCV and holdout data. Shaded channels indicate different absolute deviance of predictions. Vertical dotted lines denote commonly used SAM prevalence thresholds. Figure S14. GLM-predicted versus observed GAM (WFH + oedema) prevalence, South Sudan, by district-month, on training data, LOOCV and holdout data. Shaded channels indicate different absolute deviance of predictions. Vertical dotted lines denote commonly used GAM prevalence thresholds. Figure S15. GLM-predicted versus observed GAM (MUAC + oedema) prevalence, South Sudan, by district-month, on training data, LOOCV and holdout data. Shaded channels indicate different absolute deviance of predictions. Vertical dotted lines denote commonly used GAM prevalence thresholds. Figure S16. GLM-predicted versus observed mean WFH, South Sudan, by district-month, on training data, LOOCV and holdout data. Shaded channels indicate different absolute deviance of predictions. Vertical dotted lines denote potentially useful thresholds. Figure S17. GLM-predicted versus observed mean MUAC, South Sudan, by district-month, on training data, LOOCV and holdout data. Shaded channels indicate different absolute deviance of predictions. Vertical dotted lines denote potentially useful thresholds. Table S8. Performance of random forest models in South Sudan, by acute malnutrition outcome. Figure S18. RF-predicted versus observed GAM (WFH + oedema) prevalence, South Sudan, by district-month, on training data, LOOCV and holdout data. Shaded channels indicate different absolute deviance of predictions. Vertical dotted lines denote commonly used GAM prevalence thresholds. Figure S19. RF-predicted versus observed mean WFH, South Sudan, by district-month, on training data, LOOCV and holdout data. Shaded channels indicate different absolute deviance of predictions. Vertical dotted lines denote potentially useful thresholds.