@kieralyen
Cross-chain generalization capability evaluation of Sybil account detection models based on transaction graph features tests across Ethereum, Polkadot, and Near. Models trained on Ethereum achieve 89% accuracy when deployed on Polkadot but drop to 72% on Near due to graph structure differences. The study proposes domain adaptation techniques combining graph convolutional networks with transfer learning, improving cross-chain performance by 24%.
0 reply
0 recast
0 reaction