Optimizing Liver Disease Detection with AI: A Comparative Study of Machine Learning and Deep Learning Models

Liver disease remains a significant global health challenge, necessitating early and accurate diagnosis for effective treatment. This study proposes an AI-driven predictive framework for liver disease classification using the Indian Liver Patient Dataset (ILPD). A comparative analysis is conducted on three machine learning models—XGBoost, Support Vector Machine (SVM), and K-Nearest Neighbors (KNN)—along with a deep learning model, Artificial Neural Network (ANN). The dataset undergoes preprocessing, feature selection using SHAP (Shapley Additive Explanations), and hyperparameter tuning to enhance model performance. The results indicate that ANN achieves the highest accuracy (85.6%), followed by XGBoost (82.3%), demonstrating their superiority in predictive analytics for liver disease. SHAP analysis reveals that total bilirubin, direct bilirubin, AST levels, and albumin are the most influential biomarkers in disease classification. ROC-AUC scores confirm the robustness of ANN (0.91) and XGBoost (0.84) in distinguishing between diseased and non-diseased patients. While SVM (79.7%) and KNN (71.2%) perform reasonably well, they exhibit limitations in handling complex feature dependencies and dataset imbalances, leading to higher misclassification rates. To address misclassification issues, threshold tuning is applied, enhancing model sensitivity and ensuring higher recall for high-risk patients. The findings highlight the potential of AI-based diagnostic tools in healthcare, supporting clinicians with automated, interpretable, and efficient decision-making systems.