Machine Learning Approach for Iron Deficiency Anemia Detection in Pregnant Women Using XGBoost and CTGAN
Iron deficiency anemia (IDA) continues to pose one of the most significant challenges in maternal health, affecting nearly 40% of pregnant women worldwide according to the World Health Organization (2023). Despite advances in obstetric screening, conventional diagnostic methods such as complete blood count (CBC) tests often fail to detect early or latent stages of anemia due to physiological changes associated with pregnancy. This study introduces a robust machine learning framework integrating Extreme Gradient Boosting (XGBoost), and Conditional Tabular Generative Adversarial Networks (CTGAN) for the early detection of IDA in pregnant women. Our approach addresses the class imbalance inherent in clinical datasets and incorporates trimester-specific hematological adaptations. Using 3,944 anonymized clinical records from ASA Hospital Sarajevo (January–July 2025), we evaluated model performance across hematological features commonly used in obstetric care. The optimized model achieved a precision of 100%, recall of 65.2%, specificity of 100%, and an AUC-ROC of 0.8686. Comparative analysis against conventional CBC screening, which reached only 40.5% sensitivity, demonstrated significant improvement in detection reliability. These findings demonstrate the potential of AI-enhanced diagnostics to support early detection of IDA in pregnant women, reduce missed diagnoses, and strengthen clinical decision-making. Further multi-center validation and integration of additional biomarkers are recommended to confirm generalizability.