DEVELOPING A TRASNPARENT ANAEMIA PREDICTION MODEL EMPOWERED WITH EXPLAINABLE ARTIFICIAL INTELLIGENCE

Abstract

This paper presents a machine learning-based anaemia prediction system enhanced with explainable artificial intelligence (XAI) techniques to improve diagnostic transparency and accuracy. The study utilizes structured healthcare datasets containing demographic, nutritional, and clinical attributes to predict anaemia levels. Various supervised learning algorithms, including Random Forest, Support Vector Machine (SVM), and K-Nearest Neighbours (KNN), are implemented and compared. Feature selection and preprocessing techniques are applied to improve model performance. The proposed model integrates explainability methods such as SHAP to interpret feature contributions, enabling better clinical understanding. Experimental results demonstrate that ensemble-based models outperform traditional classifiers, achieving higher accuracy and reliability in anaemia prediction. The dataset used includes population-level health records from publicly available sources and prior cohort studies. The results indicate improved prediction performance compared to existing models, with enhanced interpretability supporting medical decision-making. This system provides a scalable and transparent approach for early detection of anaemia, particularly in resource-limited settings, contributing to improved healthcare outcomes.