On-chain behavioral signatures classify centralized exchange addresses
Original title: On-Chain Behavioral Signatures for Classifying Centralized Exchange Addresses on Ethereum Using Proof-of-Reserves
LightGBM outperforms logistic regression reaching PR-AUC 0.9594 on 3 million Ethereum addresses.
This study creates a framework for classifying Ethereum addresses of centralized exchanges using Proof-of-Reserves disclosures. Researchers extracted 58 on-chain behavioral features from 3,033,857 addresses across 15 CEXs and five non-CEX categories. Under leave-one-exchange-out evaluation, LightGBM consistently beat logistic regression with macro PR-AUC rising to 0.9594 at 730 days using 17 selected features. SHAP analysis highlighted gas behavior, ERC-20 activity and internal transactions as key signals.
- 58 on-chain behavioral features extracted
- 3,033,857 Ethereum addresses from 15 CEXs
- LightGBM macro PR-AUC 0.9594 at 730 days
- Performance robust across held-out exchanges
- Feature dataset released for follow-up research
From the original article
1 min read at arxiv.orgCentralized exchanges (CEXs) are key gateways between traditional finance and digital asset markets, yet identifying their blockchain addresses remains challenging. Existing approaches often rely on opaque attribution labels or treat exchanges as secondary classification targets, leaving generalization to unseen exchanges insufficiently examined.
By Seongchan Park · arxiv.org
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.