Developing a Transparent Anaemia Prediction Model Empowered With Explainable Artificial Intelligence

Abstract

Anaemia is a widespread global health disorder characterized by reduced haemoglobin concentration or red blood cell levels, requiring accurate and interpretable prediction for timely diagnosis and management. This study presents a transparent anaemia prediction framework developed using the Anemia Dataset from Kaggle and multiple supervised machine learning models, including Support Vector Machine, Multi-Layer Perceptron, Gradient Boosting, KNearest Neighbors, and Decision Tree. An ensemble Voting Classifier integrating Random Forest, Gradient Boosting, and Support Vector Classifier is proposed to enhance predictive robustness. Explainable Artificial Intelligence techniques, namely LIME and SHAP, are incorporated to provide featurelevel explanations and improve clinical interpretability, enabling healthcare professionals to understand model decisions. Experimental results demonstrate that the ensemble Voting Classifier achieves superior performance, attaining 100% accuracy, 100% precision, 100% recall, and 100% F1- score, outperforming individual classifiers. For practical usability, the trained model is deployed using the Flask framework, providing a lightweight and interactive web interface for real-time prediction. The system includes secure user signup and signin using an SQLite database, standardized user input processing, and clear output presentation. Based on clinical features, the application displays outcomes as “Anemia Detected” or “No Anemia Detected,” supporting reliable decision-making and trust in AI-assisted anaemia diagnosis within modern digital healthcare environments.