Bitcoin forensics dataset Elliptic++ has temporal leakage
Original title: Revisiting Elliptic++ for Bitcoin Forensics: Feature Reconstruction and Leakage-Resistant Evaluation
Re-evaluation of Elliptic++ after correcting for temporal splits and feature definitions drops random forest F1 score on illicit class from 0.846 to 0.291.
Elliptic++ is a widely used Bitcoin address-level dataset. Original positional splits violate temporal order. Feature definitions for the 55 address-level features were undocumented. Recomputing features with temporal splits based on first address appearance and recovered definitions eliminates predictive performance on illicit transactions. Random forest F1 score falls sharply and no longer outperforms logistic regression.
- F1 score drops from 0.846 to 0.291 after leakage-resistant evaluation
- Random forest no longer outperforms logistic regression baseline
- Temporal split uses only information available at prediction time
- Executable code and exact feature definitions provided for reproduction
From the original article
1 min read at arxiv.orgElliptic and its address-level extension Elliptic++ are widely used benchmark datasets in Bitcoin forensics, cited by more than 1300 academic studies according to Google Scholar. We identify future information leakage in Elliptic++. Its positional data split does not respect temporal order, and the exact definitions of its address-level features remain undocumented.
By Runlin Liang · 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.