Anomaly Detection in Blockchain-Based Metaverse Transactions Using Hybrid Autoencoder and Isolation Forest Models for Risk Identification and Behavioral Pattern Analysis

The increasing complexity of transactions within blockchain-based metaverse ecosystems has intensified the need for robust anomaly detection systems capable of identifying fraudulent, automated, or irregular behaviors. This study proposes a Hybrid Autoencoder–Isolation Forest (AE–IF) model for detecting anomalies in metaverse blockchain transactions through a combination of deep feature reconstruction and ensemble-based isolation. The proposed framework leverages the Autoencoder’s ability to learn nonlinear feature representations and the Isolation Forest’s capacity to isolate sparse anomalies, enabling the detection of both global and local irregularities. Experimental evaluation using real-world transaction data demonstrates that the hybrid model outperforms individual methods, achieving a ROC-AUC of 0.952, Precision of 0.88, Recall of 0.86, and F1-Score of 0.87. The ROC and Precision–Recall analyses confirm the model’s superior discriminative power and stability across imbalanced data distributions. Furthermore, behavioral analysis reveals distinct high-risk transaction patterns, including extended user sessions, cross-regional fund transfers, and irregular purchase behaviors. The results highlight the hybrid model’s effectiveness not only in anomaly detection but also in uncovering underlying behavioral and geographical risk factors. The proposed framework provides a scalable foundation for intelligent financial risk monitoring and cyber-fraud detection in decentralized metaverse economies.