Ensemble Machine Learning Framework for Predicting User Engagement and Risk Patterns in Metaverse Transactions

The rapid expansion of metaverse ecosystems has introduced new challenges in understanding user behavior, engagement, and financial risk within virtual transactions. This study proposes an ensemble machine learning framework that integrates LightGBM, XGBoost, and Random Forest algorithms to predict user engagement and transaction risk in metaverse environments. The model leverages temporal and behavioral features, including session duration, transaction amount, activity intensity, and short-term risk variations, to capture dynamic patterns of user interaction. Using a time-series dataset of metaverse transactions, the ensemble achieved a Mean Absolute Error (MAE) of 2.15, a Mean Squared Error (MSE) of 16.13, and an R² score of 0.9652, demonstrating exceptional predictive accuracy and generalization capability. Feature importance analysis revealed that both behavioral persistence and short-term temporal variability are critical determinants of risk. The findings highlight the effectiveness of ensemble learning for real-time risk detection, behavioral monitoring, and adaptive governance in digital economies. This study contributes to the development of intelligent, interpretable, and scalable AI-driven risk management systems for emerging metaverse platforms.