Developing a Transparent Anaemia Prediction Model Empowered With Explainable Artificial Intelligence

The worldwide health epidemic of anaemia which is a condition with low levels of red blood cells or haemoglobin requires accurate prediction models to act promptly and improve patient outcomes because it is widespread and has different causes. The effective management of anaemia is piled with obstructions, which may include the variability of diagnostic criteria, the resource limitations of healthcare, and the multifactorial nature of the disease including nutritional deficiencies, chronic disease, and genetic factors. Conventional anaemia prediction models, that predominantly rely on statistics and are trained on clinical risk scores, are frequently incapable of providing practical solutions and meaningful insights into anaemia diagnosis. There is a growing interest in focusing on Artificial Intelligence (AI) use for anaemia prediction, however, traditional AI models (black boxes) lack transparency, which causes doctors not to pick them up for practical usage. Actionable insights that are enabled by transparent AI models (white boxes) based on the explainable AI methodologies reveal the rationales of the prediction, clarify the features that are responsible for them, and help clinicians and healthcare providers. In this research work, a transparent anaemia prediction model (white box) empowered by explainable AI techniques is proposed to address the limitations of black boxes in terms of transparency. The proposed model utilizes machine learning algorithms such as Support Vector Machine (SVM), Decision Trees, K-Neighbors Classifier, and Gradient Boosting Classifier, enhanced with Explainable AI (XAI) techniques like SHAP and LIME. With the integration of explainable AI techniques like SHapley Additive exPlanations (SHAP), and Local Interpretable Model-agnostic Explanations (LIME), the proposed model offers insights into the underlying factors influencing anaemia predictions. The proposed model, significantly, represents exceptional growth in the healthcare sector and helps in bridging the gap between predictive performance and clinical interpretability, thus improving patient care and disease management strategies. The model simulation results are showing promising results in terms of the accuracy (98.13%) and the miss-rate (1.87%) which are the superior performance compared to the previous published approaches.