Similarity Report ID: oid:3618:54228609

PAPER NAME AUTHOR

Implementation-1-1.docx Tamil S

WORD COUNT CHARACTER COUNT

2023 Words 12702 Characters

PAGE COUNT FILE SIZE

8 Pages 454.1KB

SUBMISSION DATE REPORT DATE

Mar 11, 2024 12:27 PM GMT+5:30 Mar 11, 2024 12:27 PM GMT+5:30

38% Overall Similarity

The combined total of all matches, including overlapping sources, for each database.
38% Internet database

Excluded from Similarity Report

Publications database Crossref database
Crossref Posted Content database Submitted Works database

Summary
 HYBRID DEEP LEARNING APPROACH FOR
DYNAMIC ETHEREUM FRAUD DETECTION 1
Mr.E.Rajkumar, M.E., 2Mano Subash S, 3Sudharsan J, 4Balaji M
Professor1, Department of Cyber Security, SRM Valliammai Engineering College
UG Scholar2, 3, 4, Department of Cyber Security, SRM Valliammai Engineering College
[email protected], [email protected],
[email protected], [email protected]

ABSTRACT - Ethereum, a prominent block chain platform, facilitates digital asset

transactions through its decentralized network. However, the decentralized nature of Ethereum
also opens avenues for fraudulent activities, presenting challenges to network security and
integrity. This work addresses the problem of Ethereum transaction fraud by proposing an
advanced fraud detection system. Initially, it outlines the characteristics of Ethereum
transactions and the associated risks of fraudulent behavior within the network. Subsequently,
it highlights the detrimental impacts of fraudulent transactions, including financial losses and
diminished trust among users. To combat these challenges, the work presents a comprehensive
solution in the form of a Hybrid Model for Ethereum Fraud Prediction (HMEFP). The proposed
HMEFP model integrates a Customized Dense Deep Learning Model and an LSTM-based
recurrent neural network, which collectively learn intricate patterns and temporal dependencies
within Ethereum transactions. Through a fusion mechanism, the model combines the strengths
6
of both architectures, enhancing its fraud detection capabilities. Evaluation metrics such as
accuracy, precision, recall, and F1-score are employed to assess the effectiveness of the
proposed model in detecting Ethereum transaction fraud. The results demonstrate the potential
of the HMEFP to safeguard the integrity and security of the Ethereum network by accurately
identifying fraudulent transactions.
Index term: Blockchains, Feature extraction, Smart contracts, Generative adversarial
networks, Abnormal transactions,
I. INTRODUCTION
The rise of blockchain technology, particularly Ethereum, has transformed financial
transactions, enabling decentralized networks for asset transfer, smart contracts, and
decentralized applications. However, Ethereum's decentralized nature presents security
challenges, including fraudulent activities such as phishing, Ponzi schemes, and smart contract
vulnerabilities. Addressing Ethereum fraud requires innovative solutions, combining
technologies like machine learning and blockchain analysis. This study aims to develop a
comprehensive fraud detection system for Ethereum, enhancing security and trust within the
network. Through empirical analysis, the research seeks to contribute to Ethereum's integrity
as a leading blockchain platform.
II. RELATED WORKS
3
[1] Blockchain has attracted an increasing amount of researches, and there are lots of refreshing
implementations in different fields. Cryptocurrency as its representative implementation,
suffers the economic loss due to phishing scams. In our work, accounts and transactions are
treated as nodes and edges, thus detection of phishing accounts can be modeled as a node
classification problem. Correspondingly, we propose a detecting method based on Graph
Convolutional Network and autoencoder to precisely distinguish phishing accounts.
Experiments on different large-scale real-world datasets from Ethereum show that our
proposed model consistently performs promising results compared with related methods. [2]
The limitations of existing graph embedding methods in accurately capturing the characteristics
of transaction networks, particularly in dynamic environments like Ethereum. It proposes a
7
novel approach called Temporal Weighted Multidigraph Embedding (T-EDGE) to model
Ethereum transaction networks considering the dynamics and multiplicity of edges. This
3
method incorporates temporal and weighted information of edges to capture comprehensive
properties of dynamic transaction networks. Experimental results using real-world Ethereum
3
transaction data demonstrate that T-EDGE outperforms baseline embedding methods in node
classification tasks, indicating the effectiveness of considering time-dependent walks and edge
multiplicity for analyzing time-sensitive transaction networks. [3The paper addresses the
growing concern of phishing scams in the blockchain ecosystem, particularly on Ethereum,
which have become a significant threat to trading security. To combat this, the paper proposes
9
a novel approach to detect phishing scams on Ethereum by analyzing transaction records. The
method involves crawling labeled phishing addresses, reconstructing the transaction network,
3
and utilizing a network embedding algorithm called trans2vec to extract features for phishing
identification. The approach employs oneclass support vector machine (SVM) for
7
classification. Experimental results show the effectiveness of the proposed method in detecting
phishing scams on Ethereum and highlight the superiority of trans2vec in feature extraction
3
compared to existing algorithms. This work represents the first exploration of phishing
detection on Ethereum through network embedding and offers insights into feature embedding
for large-scale transaction networks. [4] The paper addresses the emerging issue of Ponzi
schemes within smart contracts, a concern as more investors engage in blockchain investment.
Existing solutions lack completeness in detecting Ponzi schemes and suffer from inefficient
detection algorithms. To address these challenges, the paper proposes an innovative approach
that incorporates bytecode features, user transaction data, and opcode frequencies to create
comprehensive features for Ponzi scheme detection. This method is coupled with an improved
LightGBM algorithm for identification. Experiments conducted on real Ethereum data
demonstrate significant improvements in accuracy, as measured by the F-score and AUC index,
compared to state-of-the-art methods. Moreover, the proposed method shows enhanced
efficiency in model training speed. Consequently, the approach offers a more accurate means
of identifying Ponzi schemes within smart contracts, ultimately reducing investment risk for
stakeholders. [5] The paper addresses the critical need for anomaly detection in smart contracts
10
to mitigate security risks such as financial fraud and money laundering, especially on the
Ethereum platform. Due to the complexity and scale of smart contract data, traditional methods
are inefficient and inadequate for extracting high-order attributes. To address this, the paper
 5
proposes Heterogeneous Graph Transformer Networks (S_HGTNs) tailored for smart contract
5
anomaly detection. This approach involves constructing a Heterogeneous Information Network
(HIN) for smart contracts and utilizing relationship matrices derived from transformer
networks to feed into convolutional networks for feature representation. The resulting node
2
embeddings are then used for classification tasks. Experimental results demonstrate the
superiority and stability of the proposed model compared to traditional methods, showcasing
its effectiveness in identifying abnormal contracts and reducing financial fraud risks.
III. MATERIAL IMPLEMENTATION
A. Data Loading and Exploring
The Ethereum fraud detection work began with the collection of Ethereum transaction
data from various sources, ensuring a diverse and representative dataset. The dataset
encompassed a wide range of transaction types, including transfers, smart contract
interactions, and token swaps. Data collection efforts aimed to capture the complexity
and diversity of transactions within the Ethereum network, providing a comprehensive
basis for fraud detection analysis.

Fig 1 – Dataset the first x number of rows when head 10

B. Data Preprocessing
Upon data collection and loading, extensive preprocessing steps were undertaken to clean
and prepare the dataset for analysis. This involved handling missing values and
standardizing data formats. Data scaling is applied to ensure uniformity across features.
 Fig 2- Turnini object variables into 'category' dtype for more computation efficiency
C. Data Splitting
The preprocessed dataset was divided into distinct training and test sets to facilitate
8
model development and evaluation. The training set comprised a majority of the data,
used to train the fraud detection model, while the test set served as an independent
dataset for model evaluation.

Fig 3- Data Splitting Testing set 30% , Training set 70%

D. Develop Defense Model

8
Feature selection plays a crucial role in enhancing the model's performance by
identifying and selecting the most relevant features from the Ethereum transaction
dataset. Our analysis reveals that Spearman correlation analysis effectively identifies
 correlated features and improves the model's predictive performance by reducing
redundancy and noise in the data.

Fig 4- Feature Selection using the Spearman Correlation Coefficient Technique

E. Feature Selection
The Hybrid Model for Ethereum Fraud Prediction (HMEFP) was constructed by
integrating a Customized Dense Deep Learning Model with an LSTM-based recurrent
neural network. This hybrid approach combined the strengths of both architectures to
effectively analyze Ethereum transaction data and detect fraudulent activities. The
dense layers in the Customized Dense Deep Learning Model captured complex
patterns within the data, while the LSTM component captured temporal dependencies
in sequential transactions, enhancing the model's fraud detection capabilities.

Fig 5- HMEFP Model Building using Hybrid - Integration of Customized Dense DL

Model & LSTM
F. Train the HMEFP Model
1
The HMEFP was trained on the designated training data using the Adam optimizer and
binary cross-entropy loss function. During training, the model iteratively adjusted its
parameters to minimize the loss function and optimize performance. Hyperparameters
such as learning rate, dropout rates, and layer sizes were fine-tuned to improve
convergence and prevent overfitting. The training process involved learning patterns
from both structured and sequential aspects of the dataset, ensuring comprehensive
fraud detection capabilities.

Fig 6- The output for HMEFP model

IV. Conclusion & Future Work

4
The proposed Hybrid Model for Ethereum Fraud Prediction (HMEFP) represents a
significant advancement in the field of fraud detection within the Ethereum network. By
integrating a Customized Dense Deep Learning Model and an LSTM-based recurrent neural
network, the HMEFP effectively captures both structured and sequential aspects of Ethereum
transaction data, allowing for comprehensive fraud detection capabilities. The customized
architecture of the dense layers and LSTM components enables the model to learn intricate
patterns and temporal dependencies inherent in Ethereum transactions. Through
experimentation and evaluation, the HMEFP demonstrates promising results in accurately
4
identifying instances of Ethereum fraud. The model's performance metrics, including accuracy,
precision, recall, and F1-score, indicate its effectiveness in distinguishing fraudulent
transactions from legitimate ones. Furthermore, the model's adaptability and robustness to
evolving fraud patterns are critical for maintaining the integrity and security of the Ethereum
network over time.
2
References

[1] Liang Chen, Jiaying Peng, Yang Liu, Jintang Li, Fenfang Xie, and Zibin Zheng. 2020.
Phishing Scams Detection in Ethereum Transaction Network. ACM Trans. Internet Technol.
21, 1, Article 10 (dec 2020), 16 pages. https://doi.org/10.1145/ 3398071
2
[2] Dan Lin, Jiajing Wu, Qi Yuan, and Zibin Zheng. 2020. T-edge: Temporal weighted
multidigraph embedding for ethereum transaction network analysis. Frontiers in Physics 8
(2020), 204.
2
[3] Jiajing Wu, Qi Yuan, Dan Lin, Wei You, Weili Chen, Chuan Chen, and Zibin Zheng. 2020.
Who are the phishers? phishing scam detection on ethereum via network embedding. IEEE
Transactions on Systems, Man, and Cybernetics: Systems (2020).

[4] Yunyi Xie, Jie Jin, Jian Zhang, Shanqing Yu, and Qi Xuan. 2021. TemporalAmount
Snapshot MultiGraph for Ethereum Transaction Tracking. In Blockchain and Trustworthy
Systems, Hong-Ning Dai, Xuanzhe Liu, Daniel Xiapu Luo, Jiang Xiao, and Xiangping Chen
(Eds.). Springer Singapore, Singapore, 133–146.
1
[5] Arnold L, Brennecke M, Camus P, Fridgen G, Guggenberger T, Radszuwill S, Rieger A,
Schweizer A, Urbach N (2019) Blockchain and initial coin offerings: blockchain’s implications
for crowdfunding. In: Business transformation through Blockchain. Palgrave Macmillan,
Cham, pp 233–272

[6] Li X, Whinston AB (2020) Analyzing cryptocurrencies. Inf Syst Front 22(1):17–22

1
[7] Liu J, Serletis A (2019) Volatility in the cryptocurrency market. Open Econ Rev 30(4):779–
811
1
[8] Mock W, Corps N (2019) CSCI49379: Intro to Blockchain (Syllabus)

[9] Malik H, Manzoor A, Ylianttila M, Liyanage M (2019) Performance Analysis of

Blockchain based Smart Grids with Ethereum and Hyperledger Implementations. In: 2019
IEEE International Conference on Advanced Networks and Telecommunications Systems
(ANTS). IEEE, pp 1–5

[10] Bartoletti M, Carta S, Cimoli T, Saia RJFGCS (2020) Dissecting Ponzi schemes on
Ethereum: identification, analysis, and impact. Futur Gener Comput Syst 102:259–277
1
[11] Chen W, Zheng Z, Ngai EC-H, Zheng P, Zhou YJIA (2019) Exploiting blockchain data
to detect smart ponzi schemes on ethereum. IEEE Access 7:37575–37586
1
[12] Aziz R, Verma C, Srivastava N (2015) A weighted-SNR feature selection from
independent component subspace for nb classifi-cation of microarray data. Int J Adv Biotec
Res 6:245–255

[13] Chen W, Zheng Z, Cui J, Ngai E, Zheng P, Zhou Y (2018) Detecting ponzi schemes on
ethereum: Towards healthier blockchain technology. In: Proceedings of the 2018 World Wide
Web Conference, pp 1409–1418
[14] Chow SS, Choo KKR, Han J (2021) Editorial for accountability and privacy issues in
blockchain and cryptocurrency. Futur Gener Comput Syst 114:647–648
1
[15] Zhang Y, Wenqiang Y, Ziyu L, Salman R, Huaihu C (2021) Detecting Ethereum Ponzi
Schemes Based on Improved LightGBM Algorithm. IEEE Transactions on Computational
Social Systems

[16] Liu L, Tsai W-T, Bhuiyan MZA, Peng H, Liu MJFGCS (2021) Blockchain-enabled fraud
discovery through abnormal smart contract detection on Ethereum. Futur Gener Comput Syst
6:133–137

[17] Hu T, Liu X, Chen T, Zhang X, Huang X, Niu W, Lu J, Zhou K, Liu Y (2021) Transaction-
based classification and detection approach for Ethereum smart contract. Inf Process Manag
58(2):102462
1
[18] Farrugia S, Ellul J, Azzopardi G (2020) Detection of illicit accounts over the Ethereum
blockchain. Exp Syst Appl 150:113318
 Similarity Report ID: oid:3618:54228609

38% Overall Similarity

Top sources found in the following databases:
38% Internet database

TOP SOURCES
The sources with the highest number of matches within the submission. Overlapping sources will not be
displayed.

link.springer.com
1 18%
Internet

arxiv.org
2 7%
Internet

researchgate.net
3 7%
Internet

ijisae.org
4 1%
Internet

journals.plos.org
5 1%
Internet

journal.uad.ac.id
6 <1%
Internet

frontiersin.org
7 <1%
Internet

dokumen.pub
8 <1%
Internet

export.arxiv.org
9 <1%
Internet

Sources overview
 Similarity Report ID: oid:3618:54228609

paperswithcode.com
10 <1%
Internet

Sources overview