Prediction of Vitamin D Deficiency Using Machine Learning, Deep Learning, and a Hybrid Model

Vitamin D deficiency is a widespread condition globally, with significant health implications. In this study, the prediction performance of Machine Learning (ML) and Deep Learning (DL) models for vitamin D deficiency was evaluated using the NHANES dataset. To enhance model performance and reduce the number of features, Principal Component Analysis (PCA) and Information Gain (IG) were employed during the feature engineering phase. The dataset was comprised of features such as age, gender, ethnicity, sun exposure time, calcium, phosphorus, and body mass index (BMI). The Synthetic Minority Over-sampling Technique (SMOTE) was applied to address class imbalance. Various models, including Decision Trees, K-NN, RNN, and LSTM, were utilized. An innovative and powerful hybrid model is provided by blending Decision Trees, XGBoost, and Artificial Neural Networks. Performance metrics such as Precision, Accuracy, Sensitivity, F1 Score, and Specifity are measured in detail. The best results were obtained with the Hybrid model (94% accuracy), while balanced performance was also shown by the KNN model (89% accuracy). The potential of ML and DL approaches in predicting vitamin D deficiency is demonstrated by this study, and an innovative approach for decision support systems in healthcare is offered through the proposed hybrid model. Another goal of this study is to provide a basis for future use and to obtain a more general version of the model by testing and demonstrating it on different data sets with examples. © 2025 IEEE.