Ensemble Machine Learning Approach for Anemia Classification Using Complete Blood Count Data

Background: Anemia is a widespread global health issue affecting millions of individuals worldwide. Early and accurate diagnosis is essential for effective treatment. Traditional diagnostic approaches rely on complete blood count (CBC) parameters, which provide valuable clinical insights but may require advanced tools to enhance diagnostic accuracy. Objective: This study aims to develop and evaluate machine learning models for classifying different anemia subtypes using CBC data. The goal is to assess the performance of individual models and ensemble methods in improving diagnostic accuracy. Methods: Five machine learning algorithms were implemented for the classification task: Decision tree, random forest, XGBoost, gradient boosting, and neural networks. In addition to evaluating individual models, ensemble techniques-including hard voting, soft voting, and stacking-were applied to enhance model performance. Results: Experimental results demonstrated that ensemble methods significantly outperformed individual models in classification accuracy. Among them, the stacking ensemble achieved the highest accuracy of 98.44%, indicating superior performance in distinguishing anemia subtypes. Conclusions: This study demonstrates that ensemble learning methods, particularly stacking, can substantially improve the performance of machine learning models in anemia classification based on CBC data. These findings suggest the potential integration of such ensemble techniques into clinical decision-support systems to assist healthcare providers in making efficient and timely diagnoses.