Hepatitis is one of the most prevalent causes of hepatocytes, the liver’s main cells, to enlarge and eventually die. Typically, hepatitis data is imbalanced. As a result, traditional statistical approaches and machine learning algorithms cannot distinguish between a few classifications. In this study, we apply machine learning algorithms to build a hepatitis B risk prediction model that improves prediction efficiency. We extracted health statistics from 155 hepatitis B individual clinical records, as well as demographic information comprising 22 factors, based on this anomalous data. We investigated various classification approaches for predicting hepatitis patients, including logistic regression, decision tree, weighted random forest (WRF) and Gaussian Naïve Bayes (GNB).In comparison to typical machine learning models, all of the algorithms listed above performed better in recognizing patients with irregular hepatitis data patterns (sensitivity >0.80). Nonetheless, LR and GNB exhibited a decreased ability to detect patients with an uncomplicated specificity of about 81%. The WRF had a higher recall (0.972), training time (0.001secods), and AUC (0.782) than the logistic regression, DT, and GNB models, indicating that it can better predict the occurrence of hepatitis. Probabilistic WRF prediction models are better suited to classifying imbalanced medical data and identifying high-risk hepatitis patients.