A Hybrid MCDM and Machine Learning Framework for Thalassemia Risk Assessment in Pregnant Women

Area of Science:

• Computational Biology and Bioinformatics
• Medical Informatics
• Public Health and Epidemiology

Background:

• Thalassemia poses a significant public health challenge in Bangladesh, particularly affecting pregnant women due to limited early screening infrastructure.
• Early identification is crucial to prevent genetic transmission and alleviate healthcare system burdens.
• Existing diagnostic methods may lack comprehensive risk stratification for targeted interventions.

Purpose of the Study:

• To develop an innovative framework for thalassemia risk assessment by integrating Multi-Criteria Decision-Making (MCDM) and machine learning (ML).
• To enhance model transparency and trustworthiness using Explainable Artificial Intelligence (XAI) techniques.
• To enable early detection and informed decision-making in maternal healthcare for thalassemia.

Main Methods:

• Integration of MCDM methods (AHP-TOPSIS) with ML algorithms (Random Forest, XGBoost, CatBoost).
• Incorporation of XAI techniques (SHAP, LIME) for model interpretability.
• Utilized real-world clinical and demographic data (16 features, 1200 samples) with rigorous feature selection and risk stratification.

Main Results:

• The XGBoost classifier, using AHP-TOPSIS-prioritized features, achieved 99.28% accuracy via 20-fold cross-validation.
• The model effectively identifies hematologic patterns indicative of thalassemia, serving as an assistive diagnostic tool.
• Explainability analyses confirmed the model's clinical transparency and reliability.

Conclusions:

• The proposed MCDM-ML framework shows significant potential for improving thalassemia risk assessment and early detection in maternal health.
• This proof-of-concept system demonstrates the feasibility of integrating MCDM, ML, and XAI for thalassemia assessment.
• Further external validation and inclusion of causal predictors are recommended before clinical deployment.