How Science Identifies Exchange Wallets on Ethereum

Have you ever wondered how researchers figure out which Ethereum wallet belongs to a big exchange? Blockchains are open to the public, but figuring out who actually owns a specific wallet address is surprisingly tricky. I recently read a study that uses machine learning to spot exchange wallets based on their behavior. It turns out centralized exchanges leave very specific clues whenever they move crypto around.

The researchers set out to accurately classify centralized exchange addresses, especially when testing on unseen exchanges. To do this, they gathered Proof-of-Reserves data covering 3,033,857 Ethereum addresses from 15 exchanges and five non-exchange categories. They extracted 58 behavioral features and compared a LightGBM model against logistic regression across lookback windows between 30 and 730 days. By using 17 selected features at 730 days, LightGBM reached a macro PR-AUC score of 0.9594 compared to 0.7697 for logistic regression. Analysis showed that gas behavior, ERC-20 token activity and internal transactions were the strongest clues. These results show that on-chain screening can be both practical and reliable for blockchain monitoring.

The study behind all this is On-Chain Behavioral Signatures for Classifying Centralized Exchange Addresses on Ethereum Using Proof-of-Reserves by Seongchan Park, Hyunho Jeon and Byungsoo Kang, with several coauthors. It lives at arxiv.org/abs/2610.41888 if you feel like reading more. I only tried to make the abstract easier to digest.

This is not financial advice. The information provided is for educational and informational purposes only.

The cover image was generated by AI.