# Exploring Unconfirmed Transactions for Effective Bitcoin Address Clustering

**Authors:** Kai Wang, Maike Tong, Changhao Wu, Jun Pang, Chen Chen, Xiapu Luo, Weili Han  
**arXiv:** [2303.01012](https://arxiv.org/abs/2303.01012)  
**Date:** 2023-03-02  
**Summary page:** https://www.summarizepaper.com/en/paper/2303.01012v2/

## Key points

- Clustering heuristics have shown that Bitcoin is not entirely anonymous
- Existing clustering methods only consider confirmed transactions in the blockchain
- Unconfirmed transactions in the mempool have yet to be utilized to improve clustering heuristics
- A data collection system is presented for capturing unconfirmed transactions
- Two case studies demonstrate user behaviors present in unconfirmed transactions but not confirmed ones
- State-of-the-art clustering heuristics are applied to both confirmed and unconfirmed transactions, reducing entities by 2.3%
- Three novel clustering heuristics are proposed to capture specific behavior patterns in unconfirmed transactions, further reducing entities by 9.8%
- Utilizing unconfirmed transactions is important for address clustering and highlights limitations of anonymity in cryptocurrencies
- This paper is the first to apply unconfirmed Bitcoin transactions for address clustering purposes

## Summary

The development of clustering heuristics has revealed that Bitcoin is not entirely anonymous, as existing clustering methods only consider confirmed transactions recorded in the blockchain. However, unconfirmed transactions in the mempool have yet to be utilized to improve the performance of these heuristics. In this paper, a data collection system is presented for capturing unconfirmed transactions and two case studies demonstrate user behaviors present in unconfirmed transactions but not confirmed ones. The state-of-the-art clustering heuristics are then applied to both confirmed and unconfirmed transactions, reducing the number of entities by 2.3% after applying co-spend heuristics on confirmed transactions alone. Three novel clustering heuristics are proposed to capture specific behavior patterns in unconfirmed transactions, further reducing the number of entities by 9.8% after applying co-spend heuristics on both confirmed and unconfirmed transactions combined. These results highlight the importance of utilizing unconfirmed transactions for address clustering and provide insight into the limitations of anonymity in cryptocurrencies. This paper is also noted as being the first to apply unconfirmed Bitcoin transactions for address clustering purposes.