Ahmed Abbas Jasim Al‐Hchaimi, M. A. Khalifa, Walid El‐Shafai
ABSTRACT Blockchain networks now support billions of dollars in daily transactions, making reliable and transparent fraud detection essential for maintaining user trust and financial stability. Yet, real‐world blockchain datasets are extremely imbalanced, with fraudulent activity representing less than 1% of all transactions. This imbalance causes conventional machine learning models to achieve deceptively high accuracy while still failing to detect a substantial portion of fraudulent events. To address this challenge, this study evaluates the performance and explainability of three models‐XGBoost, LightGBM, and Decision Tree‐on the Ethereum‐based fraud detection data, in which 58% of transactions are identified as fraud. The methodology combines vast feature engineering, k‐fold cross‐validation, and assorted resampling approaches, such as Synthetic Minority Oversampling Technique (SMOTE) and Adaptive Synthetic Sampling Nearest Neighbor (ADASYN), to revise the effect of class mismatch. Accuracy, AUC, recall, precision, F1‐Score, and Matthews Correlation Coefficient(MCC) are used to measure model performance, and SHapley Additive exPlanations (SHAP) is utilized to give global and local interpretability. Experimental results show that XGBoost combined with SMOTE or ADASYN yields the strongest performance, achieving a recall over 99%, an AUC of 1.000, and a substantially improved MCC compared to training on the raw imbalanced data. LightGBM presents a favourable precision‐recall balance, and Decision Trees demonstrate significant gains after resampling, despite their simplicity. SHAP analysis reveals that log‐transformed transaction amount, merchant‐based encoding, geographic encoding, and temporal features are the primary contributors to fraud risk. These results are important in highlighting two implications: (i) the importance of dealing with extreme class imbalance, rather than choosing increasingly sophisticated approaches, and (ii) the ability to be trusted to be explained is a requirement of responsible working in both financial and blockchain settings. The research offers a pragmatic, interpretable framework on blockchain fraud detection and future directions, including sophisticated hybrid sampling, collective learning, as well as cross‐chain generalization to enhance fraud detection in distributed systems.
Smart contracts have become the backbone of decentralized ecosystems, managing billions of dollars in assets across applications ranging from Decentralized Finance (DeFi) to digital governance. Given the immutable and autonomous nature of blockchains, the security of these contracts is paramount. A single vulnerability can lead to catastrophic and irreversible financial losses. However, despite these high stakes, a significant gap exists in understanding how developers utilize exception-handling mechanisms to enforce correctness and the specific types of logic flaws that arise from their misuse. This thesis aims to enhance smart contract security through comprehensive studies, beginning with an empirical characterization of defensive programming practices, followed by a systematic analysis of associated faults, and finally, the proposal of a novel vulnerability detection framework. It consists of the following three studies. The first study focuses on the fundamental safeguards of contract logic: state-reverting state-ments (i.e., require, revert, and throw). While these statements serve as the principal mechanisms for exception handling in Solidity, there is a lack of empirical understanding regarding their prevalence and usage patterns in the wild. To address this, the study conducts the first empirical study across thousands of real-world contracts. The results reveal that these statements are pervasive, appearing even more frequently than general-purpose if statements. The analysis further demonstrates that developers primarily use these statements to perform seven types of authority verification and input validity checks. This study establishes an understanding of how developers intend to secure contract logic. The second study investigates the landscape of faults arising from the improper use of these state-reverting statements. Although developers rely on these statements for security, incorrect implementation results in subtle bugs that traditional testing often misses. To understand these failures and benchmark detection capabilities, this study constructs the first comprehensive dataset of 320 real-world faults, curated from open-source project histories and security audit reports Through manual analysis, the study derives a taxonomy of 17 distinct fault types and distills 12 common fixing strategies. A subsequent evaluation of 12 state-of-the-art security tools against this benchmark reveals an average detection rate of only 14.4%, highlighting that existing tools are ineffective at identifying these critical logic flaws. The third study addresses the limitations of existing approaches in identifying high-level semantic vulnerabilities, specifically Price Manipulation. As indicated by the second study, traditional tools struggle with logic flaws because they often lack the ability to interpret complex economic context. To bridge this gap, this study proposes PMDETECTOR, a hybrid framework designed to proactively detect price manipulation. The framework employs a three-stage pipeline to model economic semantics: (1) static taint analysis to identify potentially vulnerable paths, (2) a two-stage Large Language Model (LLM) analysis to filter effective defenses and simulate exploitation, and (3) a final static checker to validate findings. Evaluated on 73 vulnerable and 288 benign contracts, PMDETECTOR achieves up to 100% precision and 88% recall, with GPT-4o achieving a state-of-the-art F1-score of 0.91. Furthermore, in a large-scale scan of over 8,000 recently deployed contracts, it identified 4 previously unknown vulnerabilities, confirming its practical utility in securing the DeFi ecosystem. In summary, this thesis advances the field of smart contract security by bridging the gap be-tween empirical study and automated tool development. By characterizing defensive practices and investigating the limitations of existing security tools, this work paves the way for more effective detection methods. The proposed hybrid framework demonstrates that integrating static analysis with the semantic reasoning of LLMs can effectively identify complex semantic smart contract vulnerabilities, providing the community with insights and tools to safeguard decentralized applications.
Context. Machine learning approaches for smart contract vulnerability detection are typically evaluated on synthetic benchmarks of programmatically generated code snippets. Practitioner reports and recent independent evaluations indicate that automated tools continue to miss critical vulnerabilities in production audits, yet the contribution of benchmark selection to this gap remains under-examined.Objectives. This study investigates whether surface-level Solidity features that correlate with vulnerability severity in synthetic benchmarks retain their predictive validity on professionally audited contracts, and proposes a quantitative metric for assessing benchmark suitability for severity prediction research.Methods. Fifteen features were extracted identically from a 10,448-sample synthetic Solidity benchmark and DAppSCAN, a corpus of 1,646 findings from 1,199 audit reports authored by 29 firms. Feature-severity correlations were compared using Fisher r-to-z, Kolmogorov-Smirnov, and Levene tests. Logistic Regression, Random Forest, and Gradient Boosted classifiers were trained in three conditions: in-distribution synthetic, in-distribution real, and cross-distribution.Results. Mean absolute correlation was 0.228 on synthetic data versus 0.057 on real data, a fourfold gap (all p
The security issues surrounding smart contracts have garnered significant attention due to potential vulnerabilities that can lead to financial losses and a decline in trust. Despite the development of various vulnerability detection methods by researchers, existing models often suffer from low accuracy and high false positive rates. Additionally, opcode-based vulnerability detection methods frequently introduce excessive noise due to long sequences, impairing the model’s generalization capabilities. To address these challenges, this paper proposes a dual-branch vulnerability detection model, referred to as EXCT, which integrates features from both the original opcode sequences and significant opcode sequences. We employ an improved extractive summarization technique, KTextRank, to extract important opcode sequence segments. A hierarchical Transformer is utilized for global feature extraction, while Convolutional Neural Networks (CNN) are employed for local feature extraction from the original opcode sequences. Finally, we fuse the global and local features to effectively identify specific vulnerabilities within smart contract code. Experiments conducted on two real-world datasets demonstrate that our proposed approach significantly enhances performance on publicly available datasets.
With the increased usage of Bitcoin and othercryptocurrencies, there is a need to address issues related tofraud detection in cryptocurrency systems. Such issuesinclude double-spending, money laundering, and accounthacking, among others, that Bitcoin needs to guard against.However, since Bitcoin is decentralised and transactions arenot reversible, the use of central-system approaches cannotbe applied; thus, an alternative approach must be adopted.The presented project offers a viable method of usingmachine learning for Bitcoin fraud detection. The frauddetection method is real-time, using ensemble stacking,which entails combining multiple machine learning modelsto enhance prediction capabilities. Algorithms to be usedinclude Random Forest, Gradient Boosting (XGBoost,LightGBM), Support Vector Machine (SVM), LogisticRegression, and Isolation Forest. In other words, multiplealgorithms will be used to examine Bitcoin transaction data,such as amounts transacted, transaction frequency, andtransaction patterns. Ensemble stacking allows the use of thestrengths of multiple algorithms, while the real-time functionenhances the applicability of the approach. Scalability isanother critical consideration, especially considering thenumber of Bitcoin users. This is why the use of a Flaskapplication server will be necessary for user datasubmissions, visualisation, and sending fraud notifications.Evaluation will be based on accuracy, precision, recall, andF1-score.Conclusion – The proposed solution appears quiteplausible as the fraud detection through machine learning isefficient, while scalability is one of the main features of theapproach.
Decentralized finance (DeFi) platforms have experienced a rapid increase in fraudulent activities such as price manipulation, wash trading, and anomalous transaction behavior, while traditional centralized fraud detection methods remain unsuitable due to privacy and regulatory constraints. This study proposes a privacy-preserving fraud detection framework using federated learning, enabling multiple decentralized entities to collaboratively train a machine learning model without sharing raw transaction data. A real-world decentralized exchange (DEX) dataset containing over 100,000 transactions is preprocessed and enhanced through feature engineering techniques capturing swap rate deviations, transaction volume anomalies, and temporal patterns. In the absence of labeled fraud data, a heuristic-based labeling approach is employed to simulate realistic fraud scenarios. A Logistic Regression model is trained across multiple distributed client nodes, with model parameters aggregated using the Federated Averaging (FedAvg) algorithm over several communication rounds. The experimental findings show that the federated model delivers results similar to centralized methods while preserving data privacy, proving it to be an efficient solution for secure and scalable fraud detection in decentralized financial environments.
The rise of Bitcoin has revolutionized the financial landscape, but it has also opened the door to a new era of criminal activities. Criminals take advantage of the anonymity provided by Bitcoin to conduct illicit transactions and engage in fraudulent activities. To address this issue, this paper proposes a detection model using Graph Neural Networks (GNNs) to detect fraudulent activities in the complex financial systems of Bitcoin. From the GNNs, we use EvolveGCN and EvolveGGCN to compare between them and find a powerful model that can investigate the network construction of financial transactions and capture patterns and anomalies that traditional methods may miss. In the literature, there have been a limited number of studies on Bitcoin fraud detection using GNNs, especially EvolveGGCN. Therefore, in this paper, we focus on the detection of fraud in the Bitcoin network using EvolveGCN and EvolveGGCN. In addition, we used a more recent dataset called Elliptic++, which is an extension of the Elliptic Dataset. The dataset provides valuable information on the behavior and patterns of fraudulent actions in the Bitcoin network. The results show that EvolveGGCN outperforms other models in terms of precision, recall, F1 score, and micro-F1 score. With an F1-score of 0.90 and micro-F1 of 0.93 for detecting illicit transactions in the early time steps.
The rapid growth of Ethereum has enabled innovation in digital finance, smart contracts, and non-fungible tokens (NFTs) but it has also facilitated increasingly sophisticated phishing schemes. The existing fraud detection systems have limitations in addressing large-scale phishing involving multi-hop transaction patterns, rich edge metadata, extreme class imbalance, and hierarchical account organizations. This paper proposes a hierarchical multi-scale graph convolutional network (HMG-CN) that can: (1) adaptively fuse 1-3 hop convolutions to avoid over-smoothing, (2) perform edge-aware message passing conditioned on transaction attributes, (3) discover organizational structure via two-level hierarchical pooling, and (4) combine class-balanced focal loss with contrastive learning. In an experiment with a transaction graph with 805,327 nodes and 17.1 million directed edges derived from 47,123 labeled phishing addresses, HMG-CN outperformed classic machine learning (ML), graph embeddings, standard graph neural networks (GNNs), and recent phishing detectors across five different graph sizes. On 150,000 nodes, HMG-CN attained 0.943 F1 and 0.957 AUROC, exceeding the best baseline by 8.1% and 6.6%, respectively, uniquely improving its accuracy as the graph size increased. These results demonstrate that jointly modeling multi-scale structure, edges, and hierarchy yields scalable, accurate detection under low-prevalence conditions, providing a practical foundation for real-world blockchain security.
Alex J. Sinclair, Yuhan Zhang, Michael P. O’Connor, Li Wei
Graph Neural Networks (GNNs) have become the de facto standard for modeling blockchain transaction networks. However, standard GNN architectures predominantly operate under the assumption of homophily-that connected nodes share similar labels or features. This assumption catastrophically fails in the context of Ethereum phishing detection, a quintessential heterophilic learning problem where fraudsters (phishing accounts) actively connect with unlike victims (normal accounts) to facilitate theft. In this work, we argue that the "homophily bottleneck" in existing detection systems obscures critical high-frequency signals necessary for identifying illicit activity. We introduce the Heterophilic Semantic Graph Framework (HSGF), a novel architecture designed to decouple structural roles from feature smoothing. HSGF integrates a Motif-based Semantic Sampling (MSS) strategy to capture complex, directed transactional intents (e.g., dispersing, gathering, mixing) and a Heterophily-Aware Feature Fusion (HAFF) module that prevents the oversmoothing of fraudster representations into victim representations. Extensive experiments on real-world Ethereum datasets demonstrate that HSGF significantly outperforms state-of-the-art baselines, particularly in class-imbalanced scenarios, effectively breaking the ceiling imposed by traditional homophilic aggregation.
Ye Tian, Liangliang Song, Yuanyuan Ma, Yanbin Wang · 7 authors
Detecting fraudulent accounts on Ethereum is critical for securing decentralized finance (DeFi) ecosystems. Although temporal transaction dynamics offer richer behavioral signals than static graphs, existing methods struggle to jointly model the continuous, directional, and community-level nature of fraudulent fund flows. Specifically, they inadequately capture two key cues: directional-temporal transaction patterns (integrating directionality and temporal dynamics) and higher-level behavioral communities formed by accounts with similar transactional and temporal traits.To address these, we propose TimeTrace, an unsupervised graph representation learning framework for fraud detection in Ethereum transaction networks. TimeTrace models temporal behaviors via a Directed Temporal Aggregation mechanism that explicitly distinguishes incoming and outgoing flows while adaptively emphasizing recent interactions to capture pattern evolution. Building on these directional representations, TimeTrace further incorporates a differentiable clustering module to identify latent behavioral communities and encode cluster-level relational semantics. To enhance representation coherence and structural consistency, we introduce Cohesive Embedding Regularization (CER)—a graph refinement objective combining a global Laplacian term for structural consistency, an intra-cluster Laplacian term for cluster compactness, and a fidelity term to retain initial clustering information. Additionally, we construct a new phishing account benchmark with up-to-date Ethereum transaction records for realistic evaluation. Extensive experiments on three datasets demonstrate that TimeTrace outperforms state-of-the-art methods in both accuracy and efficiency, achieving F1-score improvements of 3.62\%-10.83\% and inference speeds at least as fast as the quickest baseline.
DharmaCoin is an artificial intelligence-based cyber fraud detection system that is supposed to establish a safe and moral online financial system. The suggested model enables combining blockchain security, fraud detection with the use of artificial intelligence, biometric multi-factor authentication, and secure document verification to enhance the safeguarding of digital financial crimes. It has a framework built on the values of Indian philosophies based on logic, transparency, and ethical governance and in the name of responsible and trust-worthy digital finance. DharmaCoin manages to identify transaction fraud with a 97% F1-score and forged/altering documents with 99% accuracy in using state-of-the-art technologies SatyaAI, a real-time deepfake and identity-checking tool, and RishiGuard, an anomaly financial activity detecting tool, in 200 milliseconds. The system uses a Proof-of-Stake blockchain registry to ensure unalterable records of transactions, and to increase the level of transparency within financial deals. Performance checks show that the blockchain infrastructure has the capability of handling a rate of transactions of over 1000 transactions in less than five minutes as well as be able to verify transactions in less than five seconds hence scaling to high financial volume environments. A Pilot project involving a banking partner found that a reduction of up to 40 in time to process KYC manually led to a faster approach to operations and better fraud detection. In general, DharmaCoin offers a safe, transparent, sustainable digital finance system, which is a moral and tech-centered solution to the prevention of cyber fraud and promotes the trustful financial ecosystem in accordance with the global sustainability development objectives.
The rapid growth of Ethereum blockchain transactions has led to increased vulnerability to fraudulent activities. Traditional rule-based detection systems fail to capture complex and evolving fraud patterns. This paper presents an unsupervised deep learning approach using an Autoencoder to model normal transaction behavior and detect anomalies based on reconstruction error. The model is trained on normal transactions and evaluated on a dataset of 9,841 Ethereum accounts with 45 features. The system achieves a recall of 78.7%, precision of 69%, F1-score of 0.74, and ROC-AUC of 0.70. The proposed approach demonstrates the effectiveness of unsupervised learning for fraud detection without requiring labeled datasets.
With the rapid development of financial technology and the digital economy, fraud detection in financial transactions faces increasing challenges due to complex transaction networks, temporal dependencies, and nonlinear interactions.This study proposes an RL-LGNN framework that integrates long short-term memory (LSTM) networks, graph neural networks (GNN), and reinforcement learning (RL) for fraud detection in the financial transaction settlement process.LSTM is used to encode historical transactions as temporal sequences and extract time-dependent behavioural features.GNN then models inter-node transaction relationships and captures structural information from the transaction graph.On this basis, RL is introduced to dynamically optimise the detection strategy, thereby improving model adaptability and robustness.Experimental results on both public and real-world datasets show that the proposed framework outperforms conventional methods and achieves fraud detection accuracy above 90%.The proposed framework provides an effective solution for fraud detection in financial transaction settlement.
Phishing attacks pose significant risks to the Ethereum ecosystem, comprising over 50% of Ethereum-related cybercrimes, leading to the emergence of many machine learningbased defenses.This paper introduces a comprehensive framework aimed at enhancing machine learning-based phishing detection in Ethereum transactions.The framework addresses critical aspects such as feature selection, class imbalance, model robustness, and algorithm optimization.By systematically evaluating the strengths and limitations of existing approaches, we highlight gaps in current practices, particularly in feature manipulation and unsustainable performance outcomes.Through both analytical and experimental assessments, we demonstrate the framework's ability to streamline detection techniques, improving generalization and model effectiveness.Our findings emphasize the importance of refining detection strategies to meet the evolving challenges posed by sophisticated phishing schemes in the blockchain space.
As stablecoins become increasingly prevalent in financial crimes, their usage for illicit activities has reached a scale of USD 51.3 billion. Detecting phishing activities within stablecoin transactions has emerged as a critical challenge in blockchain security. Currently, existing detection methods predominantly target mainstream cryptocurrencies like Ethereum and lack specialized models tailored to the unique transaction patterns of stablecoin networks. This paper introduces a deep learning framework, BERTSC, based on multi-modal fusion. The model integrates three core modules graph convolutional networks (GCNs), BERT semantic encoders, and soft prompt encoders to identify malicious accounts. The GCN constructs directed multi-graph representations of account interactions, incorporating multi-dimensional edge features; the BERT encoder transforms discrete transaction attributes into semantically rich continuous vector representations; the soft prompt encoder maps account interaction features into learnable prompt vectors. An innovative three-way gated dynamic fusion mechanism optimally combines the information from these sources. The fused features are then classified to predict phishing account labels, facilitating the detection of phishing scams in stablecoin transaction datasets. Experimental results on large-scale stablecoin datasets demonstrate that BERTSC outperforms baseline models, achieving improvements of 4.96%, 3.60%, and 4.23% in Precision, Recall, and F1-score, respectively. Ablation studies validate the effectiveness of each module and confirm the necessity and superiority of the three-way gating fusion mechanism. This research offers a novel technical approach for phishing detection within blockchain stablecoin ecosystems.
This paper develops computational methods for optimizing revenue recognition in machine learning platforms operating on cloud computing infrastructure. We analyze how Artificial Intelligence as a Service (AIaaS) platforms leverage distributed computing architectures, containerization technologies (Docker, Kubernetes), and microservices patterns to deliver AI capabilities, creating complex revenue recognition challenges under IFRS 15. Our research employs algorithmic analysis to examine five critical technical challenges: (1) computational resource allocation tracking across multi-tenant cloud environments, (2) real-time transaction price determination using usage metering APIs and consumption-based billing algorithms, (3) automated revenue allocation across platform components using distributed ledger technologies, (4) temporal revenue recognition optimization through event-driven architectures and streaming data processing, and (5) network effect quantification using graph algorithms and data analytics.
BPN), MATLAB, mean absolute percentage error (MAPE) In this study, we investigate Bitcoin price volatility from December 15, 2014 to January 29, 2024 using an integrated, multisource feature set and an optimization-learning pipeline that couples Taguchi orthogonal arrays with a backpropagation network (BPN) implemented in MATLAB.Publicly available market variables were prioritized and nonquantifiable exogenous shocks were not modeled; Taguchi screening identified critical predictors and simultaneously tuned control factors (network specification, hidden-neuron count, and currency inclusion), after which the BPN was trained on aligned weekly (n = 573) and monthly (n = 108) datasets to ensure cross-market comparability.Model accuracy, assessed by mean absolute percentage error (MAPE), improved substantially after Taguchi-guided selection and configuration-weekly MAPE decreased from 3.23% to 0.36% and monthly MAPE from 6.32% to 0.07%demonstrating the efficacy of the proposed optimization framework.Out-of-sample forecasts for February-April 2025 achieved predominantly sub-10% MAPE, while high-error instances were analyzed and attributed to contributing factors, yielding decision-relevant insights for practitioners and researchers.Collectively, the results show that systematic variable selection and orthogonal-array-based model design materially enhance neural forecasts of cryptocurrency prices and provide a reproducible pathway to accurate, time-efficient prediction.
Mohammed Alhayani, Wisam Hazım Gwad, Shahab Wahhab Kareem, Moustafa Fayad
The increasing use of blockchain smart contracts has introduced new security challenges, as small coding errors can lead to major financial losses. While rule-based static analyzers remain the most common detection tools, their limited adaptability often results in false positives and outdated vulnerability patterns. This study presents a comprehensive comparative analysis of machine learning (ML) and deep learning (DL) methods for smart contract vulnerability detection using the BCCC-SCsVuls-2024 benchmark dataset. Six models (Random Forest, k-Nearest Neighbors, Simple and Deep Multilayer Perceptron, and Simple and Deep one-dimensional Convolutional Neural Networks) were evaluated under a unified experimental framework combining RobustScaler normalization and Principal Component Analysis (PCA) for dimensionality reduction. Our experimental results from a five-fold cross-validation show that the Random Forest classifier achieved the best overall performance with an accuracy of 89.44% and an F1-score of 93.20%, outperforming both traditional and neural models in stability and generalization. PCA-based feature analysis revealed that opcode-level features, particularly stack and memory manipulation instructions (PUSH, DUP, SWAP, and RETURNDATASIZE), were the most influential in defining contract behavior.
<p>The rapid evolution of decentralized finance (DeFi) has brought revolutionary innovations to global financial systems; however, it has also revealed some major security vulnerabilities, especially of smart contracts. Traditional auditing methods and static analysis tools are prone to fail in identifying sophisticated threats, including reentrancy attacks, front-running, oracle manipulation, and honeypots. This review discusses the growing role of machine learning (ML) in enhancing the security of DeFi systems. It provides a comprehensive overview of modern ML-based methods related to the detection of smart contract vulnerabilities, transaction-level fraud detection, and oracle trust assessment. The paper also provides publicly available datasets, necessary toolkits, and architectural designs used for developing and testing these models. Additionally, it provides future directions like federated learning, explainable AI, real-time mempool inspection, and cross-chain intelligence sharing. While it is full of promise, the application of ML in DeFi security is plagued by issues like data scarcity, interoperability, and explainability. This paper concludes by highlighting the need for standardised benchmarks, shared data initiatives, and the integration of ML into development pipelines to deliver secure, scalable, and reliable DeFi ecosystems.</p>
Ethereum has become a significant trading platform for financial activities such as Dapps, ICOs, and DeFi. However, it has also become a hub for criminal activities such as fraud, money laundering, and illicit fundraising. The construction of fraud detection models employing machine learning techniques is currently a mainstream research direction. Nevertheless, existing studies face significant challenges, including class imbalance in data samples and a lack of model interpretability. In this content, this work proposes a novel explainable model for Ethereum illicit account detection, ETHIAD (Ethereum Illicit Account Detection). Firstly, we pre-process the dataset by ADASYN oversampling and Lasso feature selection, etc., to more efficiently achieve feature modeling of transaction structures. Then, the ETHIAD model is trained using the XGboost algorithm, with an accuracy, precision, recall, F1 score, and AUC value of 99.70%, 99.51%, 99.02%, 99.26%, and 99.45%, respectively, the model outperforms the existing SOTA model by 0.05%-1.1%. Finally, we introduce SHAP framework to analyze the key influencing factors of illicit accounts from multiple perspectives, and the conclusions strongly enhance the explainability of the model.
Topicality. Fraudulent activities on the Ethereum blockchain pose a substantial risk to decentralized finance and require capable models not only to respond to already detected abuses but also to identify suspicious accounts proactively before losses escalate. The subject of study is the application of graph and temporal neural models to the task of classifying Ethereum accounts as benign or fraudulent, considering the structural relationships between addresses and the temporal dynamics of transactions. The purpose of this article is to develop and experimentally evaluate a neural architecture based on a multilayer perceptron as a baseline component for the subsequent integration of graph and temporal mechanisms, and to analyze its performance on the open Ethereum Fraud Detection dataset, which features a high-class imbalance. The following results were obtained. A baseline deep model for binary account classification was constructed using feature preprocessing, stratified data splitting, class weight balancing, L2 regularization, Dropout, and early stopping, which enabled the achievement of an ROC AUC value of approximately 0.98 under conditions of a pronounced dominance of the safe class. A detailed analysis of the confusion matrix and the precision, recall, and F1 metrics demonstrated an acceptable trade-off between reducing false positives and minimizing the proportion of missed fraudulent accounts, which is critical for real-world financial scenarios. Conclusion. The results indicate that a properly designed baseline neural model on tabular features can ensure high-quality proactive identification of fraudulent Ethereum accounts and serve as a starting point for further integration of graph and temporal architectures aimed at improving interpretability and robustness to the evolution of malicious behavior patterns.
Vladyslav Prosolov, Oleksandr Kushnerov, Vladyslav Sokol, Ruslan Trofymenko
Topicality. Fraudulent activities on the Ethereum blockchain pose a substantial risk to decentralized finance and require capable models not only to respond to already detected abuses but also to identify suspicious accounts proactively before losses escalate. The subject of study is the application of graph and temporal neural models to the task of classifying Ethereum accounts as benign or fraudulent, considering the structural relationships between addresses and the temporal dynamics of transactions. The purpose of this article is to develop and experimentally evaluate a neural architecture based on a multilayer perceptron as a baseline component for the subsequent integration of graph and temporal mechanisms, and to analyze its performance on the open Ethereum Fraud Detection dataset, which features a high-class imbalance. The following results were obtained. A baseline deep model for binary account classification was constructed using feature preprocessing, stratified data splitting, class weight balancing, L2 regularization, Dropout, and early stopping, which enabled the achievement of an ROC AUC value of approximately 0.98 under conditions of a pronounced dominance of the safe class. A detailed analysis of the confusion matrix and the precision, recall, and F1 metrics demonstrated an acceptable trade-off between reducing false positives and minimizing the proportion of missed fraudulent accounts, which is critical for real-world financial scenarios. Conclusion. The results indicate that a properly designed baseline neural model on tabular features can ensure high-quality proactive identification of fraudulent Ethereum accounts and serve as a starting point for further integration of graph and temporal architectures aimed at improving interpretability and robustness to the evolution of malicious behavior patterns.
Hackers may create malicious solidity programs and deploy it in the Ethereum block chain. These malicious smart contracts try to attack legitimate programs by exploiting its vulnerabilities such as reentrancy, tx.origin attack, bad randomness, deligatecall and so on. This may lead to drain of the funds, denial of service and so on . Hence, it is necessary to identify and prevent the malicious smart contract before deploying it into the blockchain. In this paper, we propose an ML based malicious smart contract detection mechanism by analyzing the EVM opcodes. After balancing the opcode frequency dataset with SMOTE algorithm, we transformed opcode frequencies to the binary values (0,1) using an entropy based supervised binning method. Then, an explainable AI model is trained with the proposed binary opcode based features. From the implementations, we found that the proposed mechanism can detect 99% of malicious smart contracts with a false positive rate of only 0.01. Finally, we incorporated LIME algorithm in our classifier to justify its predictions. We found that, LIME algorithm can explain why a particular smart contract app is declared as malicious by our ML classifier based on the binary value of EVM opcodes.