Interpretable Machine Learning Models for Beta Thalassemia Prediction: An Explainable AI Approach for Smart Healthcare 5.0

Maria Abbass¹Bilal Shoaib¹abdul ahnan khan¹Dr Anas Bilal²Asaad Algarni³Raheem Sarwar^4*

• ¹green international university lahore, Lahore, Pakistan
• ²College of Information Science and Technology, Hainan Normal University, Haikou 571158, China, Haikou, China
• ³Department of Computer Sciences, Faculty of Computing and Information Technology, Northern Border University, Rafha 91911, Saudi Arabia, rafha, Saudi Arabia
• ⁴Manchester Metropolitan University, Manchester, United Kingdom

An inherited blood disorder that bounds the production of beta globin, an important protein that has a handsome contribution in the development of hemoglobin and Red Blood Cells (RBC). This protein also enables cells to carry oxygen to tissues throughout the human body. Genetic variation in hemoglobin beta gene signals the body to make beta globin chains is the cause of beta thelasemia with three major types major, intermediate and minor. There is a need of an expert system for the diagnosis of this particular disease. This study introduces an interpretable Expert system for the prediction of Beta Thelesemia incorporating Explainable AI (XAI) techniques to enhance clinical needs. Principle component Analysis (PCA) with Synthetic Minority Over-sampling Technique (SMOTE) is applied on the Beta Thalassemia Carrier (BTC) dataset 5066 patients to reduce the dimentiality and balance the output classes. Machime learning classifiers such as Neural Networks, Recurrent Neural Networks and Long Short Term Memory (LSTM) is applied and the latest one will give the 99.30% accuracy, 99.33% precision, 99.33% recall, 99.33% specificity, and 99.33% f1 score. Furthermore ensuring the models transparency and interpretability, the proposed method integrates SHapley Ad-ditive exPlanations (SHAP) and Local Interpretable Model-Agnostic Explanations (LIME), enabling both global and local interpretability of model predictions. SHAP gives us insight into important features at the global level, while LIME explains individual predictions, making the model's decisions more comprehensible for clinical applications.