Advancing Alpha-Thalassemia Carrier Screening for Better Predictions Using Explainable AI

Thalassemia is a category of inherited blood disorders, becoming a primary health threat due to its increased prevalence worldwide. Through examination and testing of Thalassemia is widely accessible to the people. However, it places a substantial financial burden on the country. Hence, developing an alternate solution for Thalassemia diagnosis is critical. Thalassemia can be either a-thalassemia or β-thalassemia based on the hemoglobin's lack of α or β parts. This research work investigates various Machine Learning (ML) algorithms, namely, Logistic Regression (LR), Decision Tree (DT), XGBoost, Random Forest (RF), and LightGBM. It also suggests an effective ML algorithm for the diagnosis of a-thalassemia. It contains two stages; the first focuses on differentiating individuals between normal and a-thalassemia. The second stage differentiates the two subtypes of a-thalassemia carriers: silent and trait carriers. It is noticed that DT gives a better accuracy of 87% than other models investigated. Furthermore, this work embeds explainable AI techniques to interpret the working of the best-performing ML algorithm to boost the trust in medical practitioners who perform thalassemia diagnosis. Further, the contribution of features towards deciding a-thalassemia is interpreted using SHapley Additive exPlanations (SHAP) and LIME (Local Interpretable Model-agnostic Explanations).