Leakage-Safe Proxy Smart Contract Detection Using Solidity Structural Graphs and Graph Neural Networks for Secure Metaverse Blockchain Applications

Main Article Content

👀 S. Aasha Nandhini
🏢 Department of Electronics and Communication Engineering, Sri Sivasubramaniya Nadar College of Engineering, Kalavakkam - 603110, India
👀 Karthick Manoj
🏢 Department of Electrical and Electronics Engineering, AMET University, Chennai, India

Upgradeable proxy smart contracts have become a fundamental component of modern blockchain systems because they enable contract functionality to evolve without changing deployed contract addresses. Their widespread adoption in decentralized applications and Metaverse platforms introduces additional architectural complexity that complicates security analysis and automated auditing. This study proposes a Graph Neural Network (GNN)-based framework for proxy smart contract detection using a Solidity structural graph representation combined with a leakage-safe evaluation strategy. Solidity source code is transformed into graph structures that preserve syntactic and structural relationships among program elements, allowing GNN models to learn architectural patterns associated with proxy contracts. Three GNN architectures, namely Graph Convolutional Network (GCN), Graph Attention Network (GAT), and GraphSAGE, are evaluated using leakage-safe grouped cross-validation to eliminate information leakage caused by duplicated source code. Experiments were conducted on a dataset containing 377 Solidity smart contracts, which were grouped into 146 unique source-code groups after duplicate removal. The experimental results demonstrate that GCN achieved the highest ranking performance with a ROC-AUC of 0.851 and a PR-AUC of 0.606, while GraphSAGE obtained the highest classification performance with an F1-score of 0.279, a Matthews Correlation Coefficient of 0.312, and a recall of 0.500. The leakage audit confirmed that no duplicated source-code groups were shared across training, validation, and testing partitions, ensuring reliable model evaluation. These findings demonstrate that structural graph representations combined with leakage-safe evaluation provide an effective and reliable approach for proxy smart contract detection, contributing to blockchain security and supporting secure smart contract auditing in Metaverse applications.

Aasha Nandhini, S., & Manoj, K. (2026). Leakage-Safe Proxy Smart Contract Detection Using Solidity Structural Graphs and Graph Neural Networks for Secure Metaverse Blockchain Applications. International Journal Research on Metaverse, 3(4), 308–330. https://doi.org/10.47738/ijrm.v3i4.89

Article Details

Section
Articles

Similar Articles

1 2 3 4 5 > >> 

You may also start an advanced similarity search for this article.