BitcoinHeist: Topological Data Analysis for Ransomware Detection on the Bitcoin Blockchain

Cuneyt Gurcan Akcora, Yitao Li, Yulia R. Gel, Murat KantarciogluarXiv 2019 · OpenReview

Introduction

Ransomware has evolved into a multi-million dollar criminal enterprise, leveraging the pseudo-anonymity of the Bitcoin blockchain to facilitate illicit payments. While the ledger is public, identifying malicious actors remains a significant challenge for law enforcement. Traditional detection methods rely on simple heuristics, such as the "co-spending" rule, which are increasingly bypassed by sophisticated attackers using mixing services and complex transaction chains. To bridge this gap, researchers have introduced BitcoinHeist, a framework that utilizes Topological Data Analysis (TDA) to identify ransomware signatures based on the structural patterns of transactions.

Methodology: Modeling the "Shape" of Money

The BitcoinHeist framework moves beyond basic statistics by modeling the blockchain as a heterogeneous graph, where both addresses and transactions serve as nodes. This approach allows for the extraction of six key graph features for every address within 24-hour windows: Income, Neighbors, Weight, Length, Count, and Loop.

Figure 1: Representation of the Bitcoin network as a heterogeneous graph of transactions (tx) and addresses (a).

The core innovation lies in the application of TDA Mapper. Unlike standard clustering algorithms, TDA Mapper analyzes the topological "shape" of high-dimensional data across multiple scales. It generates a visualization where nodes represent clusters of addresses with similar behavioral features. If an unlabelled address falls into a cluster dominated by known ransomware signatures, it is flagged with a high suspicion score.

Figure 2: Visualization of the TDA Mapper output, showing clusters of addresses with similar topological signatures.

Data and Analysis

The study utilized a comprehensive dataset spanning a decade (January 2009 to December 2018), incorporating over 24,000 malicious addresses from 27 ransomware families, including high-profile variants like CryptoLocker, Locky, and Cerber. By focusing on transactions exceeding 0.3 BTC—the typical threshold for ransom payments—the researchers were able to filter the massive Bitcoin graph into a manageable subset for analysis.

A key insight from this analysis is that ransomware families are not monolithic. Different groups exhibit distinct "topological signatures." For instance, the WannaCry family displays unique patterns in "length" and "loop" features, reflecting a specific methodology for moving and obfuscating funds compared to other families like Locky.

Key Results and Performance

The BitcoinHeist framework demonstrated significant advantages over traditional machine learning models:

  • Superior Detection: For the Locky family, the model achieved a recall of 0.900. While the precision (0.161) reflects the inherent "needle in a haystack" nature of the problem, it represents a substantial improvement over baseline heuristics that generate overwhelming numbers of false positives.
  • Predictive Power: TDA proved particularly effective at identifying emerging threats. The model successfully forecasted the appearance of 25 new ransomware families by recognizing their geometric similarities to known malicious patterns, even when training data was sparse.
  • Operational Efficiency: Through "Shape Elimination" and "Graph Elimination," the framework reduced the pool of suspicious addresses from millions to just a few thousand, making manual investigation by security analysts feasible.

Conclusion and Future Directions

BitcoinHeist demonstrates that the "shape" of money laundering provides a recognizable signature that can be captured through topological analysis. While the false positive rate remains a challenge for predicting entirely new families, the framework significantly narrows the search space for investigators. Future research will focus on integrating real-time threat intelligence and exploring how attackers might adapt their transaction patterns to evade topological detection.

Summary updated

Abstract

Proliferation of cryptocurrencies (e.g., Bitcoin) that allow pseudo-anonymous transactions, has made it easier for ransomware developers to demand ransom by encrypting sensitive user data. The recently revealed strikes of ransomware attacks have already resulted in significant economic losses and societal harm across different sectors, ranging from local governments to health care. Most modern ransomware use Bitcoin for payments. However, although Bitcoin transactions are permanently recorded and publicly available, current approaches for detecting ransomware depend only on a couple of heuristics and/or tedious information gathering steps (e.g., running ransomware to collect ransomware related Bitcoin addresses). To our knowledge, none of the previous approaches have employed advanced data analytics techniques to automatically detect ransomware related transactions and malicious Bitcoin addresses. By capitalizing on the recent advances in topological data analysis, we propose an efficient and tractable data analytics framework to automatically detect new malicious addresses in a ransomware family, given only a limited records of previous transactions. Furthermore, our proposed techniques exhibit high utility to detect the emergence of new ransomware families, that is, ransomware with no previous records of transactions. Using the existing known ransomware data sets, we show that our proposed methodology provides significant improvements in precision and recall for ransomware transaction detection, compared to existing heuristic based approaches, and can be utilized to automate ransomware detection.
BibTeX
@misc{akcora2019bitcoinheist,
  author = {Cuneyt Gurcan Akcora and Yitao Li and Yulia R. Gel and Murat Kantarcioglu},
  title = {{BitcoinHeist: Topological Data Analysis for Ransomware Detection on the Bitcoin Blockchain}},
  year = {2019},
  eprint = {1906.07852},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/1906.07852},
}
Figure