EFFECT OF ENSEMBLE SIZE AND TECHNIQUE ON PREDICTIVE ACCURACY OF HEPATITIS B OUTCOME: A SYSTEMATIC REVIEW

Abstract

Machine Learning (ML) is an essential tool in clinical decision-making, particularly for prognosis and diagnosis. This study systematically reviews literature on ML models and ensemble learning techniques for predicting Hepatitis B virus (HBV), adhering to PRISMA guidelines. Ensemble learning combines various ML models for improved accuracy, yet concerns persist regarding the number and types of models, effectiveness, and consistency of reported metrics. The review analyzed 22 studies, revealing that ensemble learning often outperformed individual models, with notable results such as 90% accuracy from [25] and AUROC values of 0.82 to 0.88 from [14]. Limitations included reliance on small datasets, predominantly two-model ensembles, and scarce use of interpretability techniques like SHAP and LIME. Despite the potential of ensemble ML approaches for enhancing HBV prediction, the research highlights the necessity for more comprehensive comparisons, diverse modeling, and better integration of interpretability methods, advocating for a scalable and robust framework for reliable ML-based HBV prediction systems.

Keyword: Hepatitis B (HBV), Ensemble Learning, Synthetic Minority Over-sampling Technique (SMOTE), SHapley Additive exPlanations (SHAP), Local Interpretable Model-agnostic Explanations (LIME),