Artificial Intelligence for Enhanced Survival Prediction in Liver Cirrhosis: A Stacking Ensemble Approach

Background and Aims: Liver cirrhosis is a significant global health concern, causing high mortality rates due to irreversible liver damage and severe complications, including liver failure and hepatocellular carcinoma. Early and accurate survival prediction is essential for identifying high-risk individuals and facilitating timely interventions. Traditional prognostic methods often lack precision, highlighting the necessity for advanced analytical approaches. Recent advancements in artificial intelligence and machine learning present promising solutions by effectively analyzing complex patient data. Methods: This study employs a stacking ensemble learning approach that integrates multiple base models—Multilayer Perceptron, Random Forest, K-Nearest Neighbors, and Support Vector Machine—with Extreme Gradient Boosting serving as the meta-learner to enhance survival prediction accuracy. We utilized the Liver Cirrhosis Stage Classification dataset from Kaggle, derived from a Mayo Clinic study on primary biliary cirrhosis (۱۹۷۴-۱۹۸۴). This dataset originally encompassed ۱۹ clinical and demographic features and ۴۱۸ samples. To improve the robustness and generalizability of the model, the dataset was manually cleaned and augmented with synthetic data, resulting in a final synthetic sample size of ۲۵,۰۰۰. Results: The stacking ensemble model outperformed individual algorithms, achieving an accuracy of ۹۹.۰%, precision of ۹۹.۰%, recall of ۹۸.۵%, F۱-score of ۹۸.۸%, Matthews correlation coefficient of ۹۸.۱%, and area under the receiver operating characteristic curve of ۹۹.۸%. These results highlight the model's superior ability to identify complex patterns and improve survival prediction in cirrhosis patients. The ensemble approach demonstrated robustness, reduced overfitting, and effectively balanced bias and variance, making it a reliable tool for clinical applications. Conclusion: The stacking ensemble approach markedly improves prediction accuracy, serving as a reliable tool for clinical decision-making in liver cirrhosis management. By leveraging the strengths of diverse algorithms and utilizing a high-quality dataset, this method offers a comprehensive framework for survival prediction, potentially enhancing patient outcomes and reducing mortality rates. Future research could explore the application of this technique in broader clinical settings and its integration with real-time patient monitoring systems.