A Stacking Ensemble Machine Learning Model for Predicting Vitamin D Deficiency in Bangladeshi Infants

ABSTRACT Despite abundant sunlight, vitamin D deficiency remains a significant health challenge in countries like Bangladesh, driven by dietary habits, socioeconomic factors and limited sun exposure. This study leverages machine learning (ML) to predict vitamin D levels in infants aged 12–24 months, using real‐world data from the International Centre for Diarrhoeal Disease Research, Bangladesh (ICDDR,B). Data preprocessing included outlier removal via the ‐score method and Min–Max normalization for variable standardization. Ten ML models, including random forest, XGBoost, gradient boosting and Gaussian regressor, were evaluated with hyperparameter tuning to optimize performance. A stacking ensemble model, combining Gaussian regressor and extra trees as base learners and XGBoost as the meta‐learner, delivered the best results (MSE: 10.971, MAE: 2.479, RMSE: 3.312, : 0.933, MSLE: 0.005, adjusted : 0.918). Evaluation metrics such as MSE, MAE, MSLE, RMSE, , and adjusted confirmed the stacking ensemble model's superior predictive accuracy. SHAP‐based interpretability analysis on the stacking ensemble identified total (total time), total (sun exposure) and UV index as the primary predictors, enhancing clinical transparency and trustworthiness of the model. This research highlights the potential of ensemble learning in healthcare and provides a foundation for early intervention strategies to address vitamin D deficiency.