Gender-Aware Prediction of Liver Disease Using Machine Learning and Clinical Laboratory Data

Abstract

Liver disease is a major health problem that may progress silently and lead to severe clinical complications if not detected early. Machine learning offers a promising approach for supporting early screening by identifying predictive patterns from clinical and biochemical patient data. This study developed an explainable gender-aware machine learning framework for liver disease prediction using demographic information and clinical biomarkers. The dataset consisted of 570 patient records after duplicate removal, including age, gender, total bilirubin, direct bilirubin, alkaline phosphatase, SGPT, SGOT, total protein, albumin, albumin/globulin ratio, and liver disease status. Several machine learning algorithms were evaluated under three experimental scenarios: original data, class-weighted learning, and SMOTENC-based oversampling. Model performance was assessed using accuracy, precision, recall, specificity, F1-score, and ROC-AUC. The experimental results showed that Gradient Boosting combined with SMOTENC achieved the best F1-score, with an accuracy of 0.7632, precision of 0.7935, recall of 0.9012, specificity of 0.4242, F1-score of 0.8439, and ROC-AUC of 0.7759. The model correctly identified 73 of 81 liver disease cases in the testing set, indicating strong sensitivity for early screening. Gender-based evaluation showed comparable F1-scores for male and female patients, with values of 0.8430 and 0.8462, respectively. Feature importance analysis identified SGOT, alkaline phosphatase, age, and direct bilirubin as the most influential predictors. These findings suggest that an explainable and gender-aware machine learning approach can support liver disease risk prediction using routinely available clinical biomarkers, although further validation using larger and more balanced datasets is required