Evaluating the Factors and Forecasting Childhood Anemia Through Machine Learning Algorithms

Anemia, characterized by insufficient hemoglobin levels, affects a significant portion of the global population, both in developed and developing nations, and is one of the most prevalent health conditions worldwide. With timely diagnosis and proper care, the risk of anemia can be reduced, potentially saving many lives. In this context, machine learning (ML) techniques can serve as valuable tools for disease diagnosis. Therefore, the objective of this study was to determine the most effective machine learning approach while considering the risk factors for childhood anemia in Bangladesh. Secondary data from the 2011 Bangladesh Demographic and Health Survey were analyzed, with both filter (chi-square test) and wrapper (Boruta algorithm) feature selection methods used to identify significant factors. The findings revealed that 52.11% of all Bangladeshi children suffered from varying degrees of anemia (mild: 29.72%, moderate: 21.60%, and severe: 0.8%). Nine key variables—children’s fathers’ education, child age, breastfeeding status, mother’s age, mother’s education, toilet type, water source, and the number of children under five years old—were found to be directly linked to anemia. Seven machine learning algorithms (KNN, NB, SVM, RF, Bagging, Gradient Boosting, and XGBoost) were compared based on model evaluation metrics, including accuracy, sensitivity, specificity, precision, Cohen’s Kappa, F1-score, and AUC. The results showed that Gradient Boosting outperformed the other algorithms with 87.46% accuracy, 85.31% sensitivity, 96.56% specificity, 95.35% precision, 0.5713 Kappa, 0.8990 F1-score, and 0.9099 AUC. Random Forest followed closely with 83.13% accuracy, 87.36% sensitivity, 83.01% specificity, 84.10% precision, 0.3601 Kappa, 0.8555 F1-score, and 0.8531 AUC. Support Vector Machine (SVM) showed 84.46% accuracy, 0.3501 Kappa, 0.8046 F1-score, and 0.8264 AUC, while XGBoost demonstrated 75.99% accuracy, 75.87% sensitivity, 76.23% specificity, 82.76% precision, 0.3319 Kappa, 0.7913 F1-score, and 0.7599 AUC. These findings suggest that machine learning techniques—especially Gradient Boosting—can be highly effective for predicting anemia in Bangladeshi children, assisting medical professionals in early detection and intervention. The results of this study are expected to guide policymakers and healthcare providers in improving patient care and advancing Bangladesh’s progress towards achieving the Sustainable Development Goal (SDG) related to health.