Despite global efforts, anemia in childhood is still a significant public health problem as it contributes to stunting of growth, impaired immunity, cognitive deficits and compromised future productivity. Early identification of at-risk children through routine screening is limited by cost, health-system capacity, and delayed laboratory access. In this paper, we utilize a publicly available Kaggle dataset which captures the factors affecting the different levels of child anemia, to present a machine learning based framework to predict such differences. The study solves the problem of requirement for reliable and data-driven anemia risk prediction using different supervised learning algorithms and their ensemble combinations. The pipeline consists of missing-value handling, categorical encoding, feature preparation and train-test split along with some class-balancing consideration and evaluate the model(s) for a comparative evaluation. We evaluated nine baseline models namely XGBoost, LightGBM, Random Forest, Decision Tree, Support Vector Machine, AdaBoost, Logistic Regression, Naive Bayes and K Nearest Neighbors. This included ensemble models using the best performing models. The results indicate that tree-based boosting algorithms yielded the strongest predictive power, with XGBoost achieving 0.9725 accuracy and a+c ensemble balanced optimum which featured the overall highest accuracy rate at 0.9729 and lowest mse score of 0.0707. The results demonstrated that the integration of ensemble machine learning can aid an early stage screening for anemia risk and help public health planners in better targeting vulnerable children with higher efficiency. While this is not intended as a clinical diagnosis, it could play a vital role as a tool among high-need populations.