Liver illnesses, especially liver cirrhosis, acute liver failure, and liver cancer, account for about 2 million annual fatalities globally. India experiences problems with the quality of the epidemiological data resources for liver illness in terms of diagnostic precision and clinical phenotyping, uniformity of reporting, and the absence of national electronic databases, much like many other developing nations. Liver failure can be prevented by detecting and treating liver issues early on. With a data imbalance, it is essential to predict the stage of liver disease. The model's novelty is its use of the hybrid SMOTE-NMA model to manage imbalanced data for both majority and minority classes. Precision, recall, f1-score, fitting time, and test accuracy are used to gauge how well the models work. With the base models LR, RC, and GNB, the proposed model HSMOTENMA had the highest recall of 0.88, 0.94, and 0.82.