Abstract

ENHANCED LIVER DISEASE PREDICTION USING MACHINE LEARNING ALGORITHMS: IMPLEMENTATION, EVALUATION, AND IMPROVEMENTS

M. Reagan*, Gore Shriraj Laxman

Abstract

This study presents an enhanced machine learning framework for liver disease prediction using the Indian Liver Patient Dataset (ILPD). Prior research primarily relied on Logistic Regression, K-Nearest Neighbours, and Support Vector Machines, achieving moderate accuracy but lacking depth in evaluation. In this work, we re-implemented these baseline models and extended the scope by incorporating Decision Tree, Random Forest, Naive Bayes, Gradient Boosting, and XGBoost. Comprehensive preprocessing steps—including label encoding, outlier treatment, normalization, and stratified traintest splitting—were applied to improve model reliability. Performance was assessed using accuracy, sensitivity, specificity, and ROC-AUC to capture clinical trade-offs between detecting diseased and healthy individuals. Results show that Logistic Regression and SVM achieve high sensitivity but poor specificity, Naive Bayes excels in specificity but underperforms in sensitivity, while ensemble methods—particularly XGBoost—offer the best balance across all metrics. Overall, the expanded approach provides a more robust and clinically meaningful evaluation framework for liver disease prediction.

Keywords: Liver Disease, Machine Learning, Classification, ROC-AUC, XGBoost.