XGBoost-based Diagnostic Model for Hepatitis B Detection using a Real-World Clinical Dataset

Hepatitis B virus (HBV) infection continues to pose a major public health challenge worldwide, particularly in regions with limited diagnostic infrastructure. Timely and accurate detection plays a crucial role in preventing disease progression to cirrhosis, liver failure, or hepatocellular carcinoma. This study proposes a machine learning-based diagnostic framework using the Extreme Gradient Boosting (XGBoost) algorithm for the detection of Hepatitis B. The framework is developed and evaluated using a primary clinical dataset collected from civil hospitals in Haryana, India. Thirteen routinely gathered demographic, symptomatic and biochemical features are present in the dataset, reflecting real-world situations. All missing values were handled, features were standardized and categorical variables were encoded to numeric values. The best parameters for the XGBoost model were found using a Randomized SearchCV with a wide space and cross-validation was used to validate the robustness of the model. A tuned version of the model performed well in diagnostics, giving mean accuracy of 83.22%, precision of 88.26%, recall of 80.62%, an F1-score of 84.24% and an area under the ROC curve (AUC) of 92.49% in comparison of existing state-of-the-art (SOTA) models. Such findings indicate that ensemble learning techniques can build useful, interpretable tools for diagnosis from local healthcare data. The planned XGBoost-based approach is a promising and affordable method for detecting Hepatitis B early, primarily where expert resources for diagnosis are low. Future efforts will focus on checking the model in unrestricted populations and developing methods that clarify its outcomes for clinical use.