This research explores the use of an ensemble comprising Random Forest, XG-Boost, and Isolation Forest algorithms for anomaly detection and risk analysis of blockchain transactions in the Open Metaverse. We created a multi-phase strategy using temporal feature engineering, risk assessment frameworks, and behavioral pattern analysis using an extensive dataset of 78,600 transactions. The ensemble model achieved 99% classification accuracy and zero false positives for valid transactions, while effectively detecting phishing (84% precision, 68% recall) and scam activities (81% precision, 91% recall). With 56.19% of the categorization, risk assessment measures were the most important predictive indicators, followed by transaction patterns and behavioral attributes. Our cross-validation analysis resulted exceptional stability in performance, with standard deviation of 0.0004 across diverse transaction scenarios. This confirms that our methodology can reliably detect intricate risk patterns, playing a crucial role in strengthening transaction security as the Open Metaverse continues to evolve.