Machine Learning-based Predictive Models for Early Diagnosis of Liver Disease

Abstract

Liver disease is a major global health issue, contributing to nearly 2 million deaths annually. Early detection is crucial, yet traditional diagnostic methods are invasive and costly. This study proposes a machine learning-based framework for liver disease diagnosis using 30,690 patient records, incorporating demographic details, liver enzyme levels, and bilirubin measurements. The methodology includes data preprocessing, feature selection, and model evaluation across 13 machine learning algorithms. Key predictive features—Total Bilirubin, Direct Bilirubin, SGPT, SGOT, and Alkaline Phosphatase— were identified using Chi-squared test, ANOVA F-value, Mutual Information, and Random Forest Importance. Among the models, Decision Tree, Bagging Classifier, and XGBoost demonstrated superior performance, achieving over 99% accuracy. The Decision Tree model exhibited the highest computational efficiency (0.0009 seconds prediction time), making it ideal for real-time clinical applications. The study underscores the potential of machine learning in non-invasive, scalable, and accurate liver disease diagnostics. Future work includes extending the model for personalized medicine and advanced liver disease subtypes.