Application and Evaluation of Ensemble Machine Learning Models for the Differential Diagnosis of Clinical and Hematological Syndromes in the Health Passport Software Package
Abstract
Anemia remains a widespread clinical and public health challenge, particularly in women and children, often leading to delayed or incorrect treatment due to diagnostic complexity. While traditional diagnosis relies on manual interpretation of laboratory data, recent advances in machine learning offer promising automated alternatives. This study proposes a novel diagnostic framework integrating ensemble and neural network models trained on a comprehensive clinical dataset comprising over 30 hematological and biochemical features-including hemoglobin, red blood cell count, hematocrit, red cell distribution width, serum B12, and iron levels. The approach systematically compares baseline classifiers (SVM, Logistic Regression, Decision Trees) with advanced ensemble techniques (Random Forest, Bagging, Voting, XGBoost, LightGBM, Stacking) and integrates them into the Health Passport software package. Experimental validation shows that ensemble models significantly improve diagnostic precision, with XGBoost achieving over 98% accuracy and neural networks providing probabilistic outputs for anemia subtypes. These results confirm the efficacy of intelligent diagnostic systems in enhancing clinical decision-making and supporting personalized medicine in hematology.