Globally fatty liver disease has become a fundamental cause of liver disease. As per epidemiological studies, the fatty liver encompasses 9 to 32% of the general population in India, with the highest prevalence among those who are bulky or obese, and also those who have suffered from diabetes. As a method for computational image analysis, deep learning is gaining popularity. By using high-dimensional image-derived features to perform diagnostic or predictive tasks, liver imaging will be able to extend its capabilities beyond what is traditionally possible with visual image processing. In order to determine the stages of fatty liver, the liver volume should be evaluated. The proposed approach classifies the segmented liver as normal and abnormal. Different MLAs are compared Naïve Bayes (NB), Random Forest (RF), and Decision Tree (DT). Based on this accuracy, the NB classifier delivers a unique finer output for classifying normal and abnormal liver. The accuracy rates are 99.9% for NB, 98% for RF and 97.4% for DT. From these results, the NB provides better accuracy for classification.