Machine Learning-Based Multi-Class Classification of Beta-Thalassemia Trait and Hemoglobin E Carrier Status Using HPLC and Complete Blood Count Parameters

Thalassemia and hemoglobin E (HbE) disorders are a major public health burden across South and Southeast Asia, where Bangladesh and neighboring West Bengal, India, record carrier prevalences of 10 to 13 percent. Accurate discrimination between beta-thalassemia trait (BTT), HbE carrier status, and the normal hemoglobin pattern is essential for antenatal and premarital screening, yet existing machine learning studies were developed mainly on populations where beta-thalassemia predominates, leaving a gap for HbE-prevalent South Asian populations. We applied six supervised machine learning algorithms (Logistic Regression, K-Nearest Neighbors, Support Vector Machine, Random Forest, XGBoost, and LightGBM) to a publicly available dataset of 12,885 individuals screened in West Bengal, India, comprising 11,777 normal individuals (91.4 percent), 562 BTT carriers (4.4 percent), and 546 HbE carriers (4.2 percent). Features included high-performance liquid chromatography (HPLC) hemoglobin fractions (HbA0, HbA2, HbF) and complete blood count (CBC) parameters (RBC, hemoglobin, MCV, MCH, MCHC, RDW-CV), with age and sex. Class imbalance was addressed using the Synthetic Minority Oversampling Technique applied only to the training set. Performance was evaluated on a held-out 20 percent test set and validated by 10-fold cross-validation, and SHapley Additive exPlanations (SHAP) were computed for interpretability. Random Forest and LightGBM achieved perfect classification (accuracy 100 percent, macro F1 100 percent, area under the curve 1.000); XGBoost reached 99.96 percent accuracy with 10-fold cross-validation confirming 99.98 percent. A CBC-only analysis achieved 92.01 percent accuracy with XGBoost, supporting use as a low-resource triage tool. SHAP identified HbA2 as the dominant predictor for all classes and MCH as the most informative CBC parameter for BTT detection. Ensemble machine learning can accurately classify BTT, HbE carrier, and normal hemoglobin patterns with zero inter-carrier misclassification, supporting integration into South Asian screening workflows.