Ensuring security in edge computing through effective blockchain node detection

Shenqiang Wang, Zhaowei Liu, Haiyang Wang, Jianping WangJournal of Cloud Computing Advances Systems and Applications 2023 · DOI

Abstract

The rapid development of blockchain technology has garnered increasing attention, particularly in the field of edge computing, due to its potential for data privacy protection. However, security threats persist; attackers may manipulate nodes to tamper with transaction records or deploy Sybil attacks to compromise network reliability. This paper proposes $T^2A2vec$, a blockchain node detection method designed to provide a secure and credible solution. By constructing a transaction graph from Ethereum data and employing a multi-faceted feature extraction process—using BP neural networks for account features and a biased random walk strategy for transaction features (based on time, type, and amount)—the method achieves high detection accuracy. The fused representations are then processed by classifiers to identify malicious nodes effectively.

Introduction

Edge computing decentralizes data processing by moving it closer to the network edge, which enhances performance but introduces significant security vulnerabilities. While blockchain is often integrated to secure these environments, the blockchain layer itself is a target for manipulation. Malicious actors can create fake nodes or compromise existing ones to distort transaction history. Traditional detection methods often fail due to data bias—using datasets that lack temporal or spatial representativeness—and feature sparsity, where critical transaction metadata like timing, value, and type are ignored.

The $T^2A2vec$ Framework

To address these limitations, the researchers developed the $T^2A2vec$ (Transaction Time, Type, and Amount) framework. It follows a comprehensive pipeline for node representation and classification.

The $T^2A2vec$ framework overview

The framework begins by modeling the blockchain as a multi-directed graph where accounts are nodes and transactions are edges. A Backpropagation (BP) neural network is utilized to extract 18 statistical features for each account, including balance and transaction frequency. Simultaneously, the transaction graph is processed to extract structural features through specialized embedding techniques.

Biased Random Walk Strategies

The core innovation of $T^2A2vec$ lies in its biased random walk strategy. Unlike standard graph embedding methods that treat all edges as equal, $T^2A2vec$ incorporates the semantic context of transactions to capture the "intent" behind the data:

  • Time-based: Biases walks toward recent activity, capturing the rapid transaction patterns often seen before illicit nodes cash out.
  • Type-based: Assigns weights based on the nature of the transaction, such as simple transfers, contract creation, or contract calls.
  • Amount-based: Prioritizes larger transactions, which typically indicate more significant node relationships.
Blockchain transaction graph and edge types

These graph-based embeddings are fused with statistical account features to create a robust vector for classification. This final representation is fed into a classifier, such as Random Forest, to label the node as "normal" or "illicit."

Experimental Results

Evaluated using Ethereum transaction data as a proxy for edge computing scenarios, $T^2A2vec$ demonstrated superior performance. On a dataset of approximately 10,000 nodes and 1.1 million transactions, the model achieved a precision of 93.21% and an F1-score of 93.08% using a Random Forest classifier, significantly outperforming standard Graph Neural Networks like GCN and GraphSAGE.

The research identified a distinct "illicit profile": malicious nodes tend to have lower degrees of connectivity and show a marked asymmetry in transaction flow, often sending significantly more funds than they receive. Once a node is flagged as malicious, ordinary users stop interacting with it, effectively isolating it in the graph.

Conclusion

$T^2A2vec$ offers a powerful tool for enhancing security in blockchain-enabled edge computing. By integrating temporal, structural, and statistical data, it provides a nuanced detection mechanism for malicious nodes. Future research will explore multi-class detection to categorize specific illicit behaviors and validate the model on native edge computing blockchain platforms as they become more prevalent.

Summary updated

Abstract

Abstract The rapid development of blockchain technology has garnered increasing attention, particularly in the field of edge computing. It has become a significant subject of research in this area due to its ability to protect the privacy of data. Despite the advantages that blockchain technology offers, there are also security threats that must be addressed. Attackers may manipulate certain nodes in the blockchain network, which can result in tampering with transaction records or other malicious activities. Moreover, the creation of a large number of false nodes can be utilized to gain control and manipulate transaction records of the blockchain network, which can compromise the reliability and security of edge computing. This paper proposes a blockchain node detection method named $$T^2A2vec$$ <mml:math xmlns:mml="http://www.w3.org/1998/Math/MathML"> <mml:mrow> <mml:msup> <mml:mi>T</mml:mi> <mml:mn>2</mml:mn> </mml:msup> <mml:mi>A</mml:mi> <mml:mn>2</mml:mn> <mml:mi>v</mml:mi> <mml:mi>e</mml:mi> <mml:mi>c</mml:mi> </mml:mrow> </mml:math> that provides a more secure, credible, and reliable solution to address these challenges. In order to achieve $$T^2A2vec$$ <mml:math xmlns:mml="http://www.w3.org/1998/Math/MathML"> <mml:mrow> <mml:msup> <mml:mi>T</mml:mi> <mml:mn>2</mml:mn> </mml:msup> <mml:mi>A</mml:mi> <mml:mn>2</mml:mn> <mml:mi>v</mml:mi> <mml:mi>e</mml:mi> <mml:mi>c</mml:mi> </mml:mrow> </mml:math> , a transaction dataset that is evenly distributed in both space and time was collected. The transaction dataset is constructed as a transaction graph, where nodes represent accounts and edges describe transactions. BP neural network is used to extract account features, and a random walk strategy based on transaction time, type, and amount is used to extract transaction features. The obtained account features and transaction features are fused to obtain account representation. Finally, the obtained node representation is fed into different classifiers to identify malicious nodes.
BibTeX
@article{wang2023ensuring,
  author = {Shenqiang Wang and Zhaowei Liu and Haiyang Wang and Jianping Wang},
  title = {{Ensuring security in edge computing through effective blockchain node detection}},
  journal = {Journal of Cloud Computing Advances Systems and Applications},
  year = {2023},
  doi = {10.1186/s13677-023-00466-y},
}
Figure