On-chain features identify centralized exchange Ethereum addresses accurately
On-Chain Behavioral Signatures for Classifying Centralized Exchange Addresses on Ethereum Using Proof-of-Reserves
Computers and SocietyCryptography and SecurityMachine Learning
Summary
It is hard to figure out which Ethereum blockchain addresses belong to big centralized cryptocurrency exchanges. The authors created a method to spot these addresses by looking at certain patterns in how they behave on the blockchain, using public data about their reserves. They tested their method on millions of addresses from many exchanges and found that a machine learning model called LightGBM worked well, especially when focusing on key features like transaction gas use and token activity. This approach helps blockchain watchers identify exchange addresses more clearly and can be used and improved by others.
What this means in practice
- •For blockchain compliance teams: Detect Ethereum addresses of centralized exchanges for better regulatory monitoring using on-chain behavioral patterns.
- •For crypto analytics providers: Improve exchange labeling in Ethereum datasets using machine learning features derived from Proof-of-Reserves disclosures.$Commercial implications: Enables creation of enhanced compliance and market analysis tools that accurately identify exchange activity on Ethereum.
Authors
Seongchan Park, Hyunho Jeon, Byungsoo Kang, Gaeun Kim, Minseob So, Seongyong Kim
Abstract
Centralized 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. This study develops a classification framework using Proof-of-Reserves (PoR) disclosures, with 3,033,857 Ethereum addresses from 15 CEXs and five Non-CEX categories. We derive 58 on-chain behavioral features and evaluate LightGBM against logistic regression (LR) under a leave-one-exchange-out protocol. Across five lookback windows (30-730 days), LightGBM consistently outperformed LR, with macro PR-AUC increasing from 0.6611 to 0.9594 using the selected 17 features, compared with 0.7697 for LR at 730 days. Robustness checks showed comparable performance when feature selection and hyperparameter optimization were repeated across splits and held-out exchanges, while exchange-balanced reweighting improved macro PR-AUC. SHAP analysis and feature-group ablation identified gas behavior, ERC-20 token activity, and internal transactions as key predictive signals in the constructed dataset. These findings support interpretable CEX address screening for blockchain monitoring, while the released feature dataset enables reproducible follow-up research.