Comparative Analysis of LightGBM and XGBoost for Predictive Risk Assessment in Blockchain Transactions within the Metaverse

The growing integration of blockchain technology within the metaverse has created an urgent need for effective systems to assess and mitigate transaction risks. This study investigates the use of machine learning models, specifically LightGBM and XGBoost, for predictive risk analysis in blockchain transactions. A dataset comprising 50,000 blockchain transactions, with 75% categorized as low-risk and 25% as high-risk, was used to evaluate the performance of these models across key metrics. LightGBM emerged as the superior model, achieving an accuracy of 91.2%, surpassing XGBoost's 89.5%. Additionally, LightGBM recorded an AUC-ROC score of 0.94, outperforming XGBoost’s 0.92. In terms of computational efficiency, LightGBM demonstrated clear advantages. It required only 80 seconds for training and 10 milliseconds per prediction, whereas XGBoost needed 120 seconds for training and 15 milliseconds for prediction. Feature importance analysis further highlighted the pivotal role of the Risk Score, which contributed 40% and 35% to the predictive power of LightGBM and XGBoost, respectively. Other significant features included Amount (USD) and Session Duration, showcasing the relevance of both behavioral and transactional data in risk prediction. These results underscore LightGBM's suitability for real-time risk assessment, making it a reliable and efficient tool for managing large transaction volumes in blockchain ecosystems. However, this study also acknowledges some limitations, including the imbalanced dataset and the static nature of the models, which may struggle with evolving transaction patterns. Future research could address these challenges by employing advanced resampling techniques to balance the dataset, incorporating additional contextual features, and developing adaptive models capable of handling dynamic risk profiles. Through these advancements, this research contributes to the foundation for scalable and secure risk assessment systems, fostering trust in blockchain-based metaverse applications.