XAI-Driven Comparative Framework for Anemia Diagnosis Using SHAP and LIME

Abstract

Anemia is a widespread health problem, particularly among young children, pregnant or postpartum women, and menstruating girls and women. It arises from a lack of healthy red blood cells to carry oxygen, leading to fatigue, shortness of breath, and reduced physical capacity. Often tied to poor nutrition or underlying conditions in low-resource areas, anemia demands early and accurate detection for effective care. While machine learning (ML) has shown strong potential for improving diagnostic accuracy, it lack’s of transparency often referred to as the black box problem can make it difficult for clinicians to fully trust and adopt these systems in practice. To improve model interpretability, two widely used explanation methods, SHAP (SHapley Additive exPlanations) and LIME (Local Interpretable Model-agnostic Explanations), were integrated with the ML classifiers. A publicly available dataset of 364 patient blood test results was utilized, and a supervised oversampling hybrid resampling technique was applied to address class imbalance in the multi-class classification task. The results are showed that the LightGBM model achieved the best performance, with an accuracy of 98.84% and a perfect ROC-AUC score of 1.00. It was great at telling different anemia types apart, which helps with treatment choices. By integrating SHAP and LIME, the framework delivers patient-specific explanations that make the model’s decisions easier to understand and more practical for clinical use. With its high accuracy and clear interpretability, this approach offers a reliable solution for anemia diagnosis, particularly in resource-limited settings, and can help encourage wider adoption of intelligent healthcare systems.

Ähnliche Arbeiten