Non-alcoholic fatty liver disease (NAFLD) is a prevalent global health concern strongly linked to obesity and metabolic syndrome (MS). Traditional diagnostic methods are invasive and costly, highlighting the need for non-invasive approaches. ML algorithms offer promise in analyzing diverse datasets to develop accurate prediction models. The paper discusses various ML algorithms, feature selection methods, and evaluation techniques used in NAFLD prediction. It also presents a comparative analysis of classification algorithms’ diagnostic outcomes, demonstrating random forests with 93% accuracy and 86% area under the receiver operating characteristic (ROC) curve as the most accurate classifier. The study underscores the importance of automating NAFLD diagnosis and improving model performance, emphasizing the potential for ML algorithms in clinical practice for risk assessment and personalized management of NAFLD.