Hybrid Ensemble Learning for Anomaly Detection in Metaverse Transactions Using Isolation Forest, Autoencoder, and XGBoost

The rapid expansion of metaverse platforms has increased the volume and complexity of digital transactions, creating a greater need for reliable anomaly detection systems. This study proposes a hybrid ensemble learning framework that integrates Isolation Forest, Autoencoder, and XGBoost using a meta learning approach to detect anomalous transactions in metaverse environments. The framework combines unsupervised and supervised learning to identify structural irregularities, behavioral deviations, and contextual patterns associated with high-risk activities. Using a transaction dataset containing behavioral, contextual, and numerical features, the hybrid model was evaluated against its individual components. The results show that the proposed framework achieves superior accuracy, precision, recall, and ROC AUC values compared to standalone models. The analysis of feature importance indicates that quantitative variables, including transaction amount, session duration, and risk score, provide the strongest predictive contribution, while contextual and behavioral factors improve model interpretability and generalization. Principal Component Analysis further visualizes the separation between normal and anomalous clusters, confirming that the hybrid ensemble effectively captures latent relationships within high-dimensional transaction data. Overall, the findings demonstrate that the proposed approach provides a robust and scalable solution for detecting irregular patterns in metaverse-based blockchain transactions. This model also offers practical implications for real-time financial risk assessment and digital security management in decentralized virtual economies.