A HYBRID ENSEMBLE LEARNING FRAMEWORK FOR ACCURATE HEPATITIS DISEASE CLASSIFICATION

Abstract

Hepatitis, a severe inflammatory condition of the liver, continues to pose substantial diagnostic challenges due to its heterogeneous clinical manifestations and overlapping symptoms with other hepatic disorders. Early and accurate detection is crucial for timely intervention, effective treatment planning, and reducing the risk of progression to chronic liver disease or liver failure. Conventional machine learning models, although widely applied in medical diagnosis, often struggle to achieve high reliability due to limitations such as overfitting, feature noise sensitivity, and inability to fully capture complex nonlinear relationships present in clinical datasets. Hepatitis remains a major global health burden, requiring reliable early diagnostic support to improve clinical outcomes. Conventional single-model approaches suffer from variability in performance due to dataset imbalance, missing values, and overlapping clinical features. This paper proposes a Hybrid Ensemble Learning Framework (HELF) combining Gradient Boosting, Random Forest, and Logistic Regression through a stacked metalearner to enhance predictive accuracy for Hepatitis disease classification. The framework applies advanced preprocessing including missing-value imputation using K-Nearest Neighbours (KNN), Recursive Feature Elimination (RFE), and class rebalancing through SMOTE. Experimental evaluation using the UCI Hepatitis dataset demonstrates significant performance improvement, achieving 96.42% accuracy, 95.71% F1-score, and 0.98 AUC, outperforming baseline single models. The framework provides a reproducible, clinically feasible solution for augmenting earlystage hepatitis risk assessment.