Liver cirrhosis is a serious, progressive disease with high morbidity and mortality rates, especially in patients with limited access to healthcare. Early prediction of cirrhosis outcomes can enable timely intervention and improve patient survival rates. In this study, a robust machine learning model was developed to predict the survival outcomes of cirrhosis patients using a novel dataset from the Mayo Clinic, which includes demographic, clinical, and biochemical data. The outcomes were classified into three categories: censored (C), alive post-liver transplant (CL), and deceased (D). A range of machine learning models, including XGBoost, Random Forest, and LightGBM, were evaluated using accuracy, F1-score, and ROC-AUC metrics. XGBoost outperformed other models, achieving the highest F1-scores for the censored and deceased classes, although performance on the alive class was lower due to class imbalance. The study highlights the effectiveness of ensemble methods in handling complex, imbalanced datasets; with a primary focus on F1 score for a multiclass classification task and underscores the need for improved techniques for minority class prediction.