Unlocking insights: Using machine learning to identify wasting and risk factors in Egyptian children under 5

• Published in: Nutrition (Burbank, Los Angeles County, Calif.) Los Angeles County, Calif.), 2025-03, Vol.131, p.112631, Article 112631
• Main Authors: Hendy, Abdelaziz, Abdelaliem, Sally Mohammed Farghaly, Sultan, Hosny Maher, Alahmedi, Shorok Hamed, Ibrahim, Rasha Kadri, Abdelrazek, Eman Mohamed Ebrahim, Elmahdy, Masani Abdelbagi Ahmed, Hendy, Ahmed
• Format: Article
• Language: English
• Subjects: child health; Child nutrition; data collection; decision support systems; Egypt; Machine learning; malnutrition; nutritional status; Public health; regression analysis; risk; socioeconomics; Wasting; Weight-for-height Z-scores
• ISSN: 0899-9007; 1873-1244; 1873-1244
• DOI: 10.1016/j.nut.2024.112631

•Machine learning models effectively identified key determinants of wasting in Egyptian children.•XGBoost outperformed other classifiers with a 94.8% accuracy and 99.4% ROC-AUC score.•Maternal BMI, child weight, and regional disparities were significant predictors of wasting.•Rural areas showed higher prevalence of wasting due to food insecurity and limited healthcare.•76.2 % of the children studied had a normal nutritional status, 5.2% of the children suffered from wasting, with 1.7% experiencing severe wasting. Malnutrition, particularly wasting, continues to be a significant public health issue among children under five years in Egypt. Despite global advancements in child health, the prevalence of wasting remains a critical concern. This study employs machine learning techniques to identify and analyze the determinants of wasting in this population. To evaluate the prevalence of wasting among children under five years in Egypt and identify key factors associated with wasting using machine learning models. This study is based on secondary data sourced from the Demographic and Health Surveys (DHS), conducted in 2005, 2008, and 2014. Six machine learning classifiers (XGBoost, Logistic Regression, Random Forest, Gradient Boosting, K-Nearest Neighbor, and Decision Tree) were applied to the dataset. The study included children under five years of age, focusing on nutritional status, maternal health, and socio-economic factors. The dataset was cleaned, preprocessed, encoded using one-hot encoding, and split into training (70%) and test (30%) sets. Additionally, k-fold cross-validation and the StandardScaler function from Scikit-learn were used. Performance metrics such as accuracy, precision, recall, F1 score, and ROC-AUC were used to evaluate and compare the algorithms. It was observed that 76.2% of the children in the dataset have normal nutritional status. Furthermore, 5.2% were found to be suffering from wasting (1.7% experiencing severe wasting and 3.5% moderate wasting), with notable regional disparities. The XGBoost model outperformed other models. Its efficiency metrics include an accuracy of 94.8%, precision of 94.7%, recall of 94.7%, F1 score of 94.7%, and an ROC-AUC of 99.4%. These results indicate that XGBoost was highly effective in predicting wasting. Machine learning techniques, particularly XGBoost, show significant potential for improving the classification of nutritional status and addressing wasting among children in Egypt. However, the limi