Detecting anomalous transactions in the metaverse is a major problem that has serious security implications. The review of available studies on machine learning models used to detect these anomalies is also scanty. To achieve this, four machine learning approaches (i.e., Decision Trees, Random Forest, Support Vector Classification (SVC), and K-Nearest Neighbors (KNN)) will be evaluated and com- pared in terms of their ability to determine the most effective model for anomaly detection. The dataset contains 78,600 transaction records with fields such as timestamps, addresses, amounts, and risk classifications. This comparative analysis, which considers ac- curacy, precision, F1 score, and recall as evaluation metrics, reveals that Random Forest outperforms the other models. The findings of this study are expected to provide insights for building safer and more secure metaverse environments.