It is concluded that each component of a layered fraud system contributes only under specific conditions, and that a plausible rationale from an investigation agent is not evidence of a better decision.
Abstract
Fraud detection systems must scale with rising transaction volume while remaining explainable and reviewable. We study a layered pipeline on the PaySim dataset that combines a gradient-boosted classifier, graph-derived structural features, an autoencoder-based anomaly signal, TreeSHAP explanations, and a bounded LLM investigation agent applied to cases the classifier scores uncertainly. Before any model comparison, we identify and remove a simulator-specific balance shortcut that would otherwise inflate baseline performance. After this correction, neither the graph features nor the anomaly signal improves Average Precision on the full test set. Both, however, rank fraud better within the subset of cases receiving intermediate baseline scores. In a controlled experiment with injected multi-account fraud rings, engineered structural features recover all injected test transactions, while the tabular baseline misses roughly a quarter of them. The investigation agent underperforms direct thresholding of the classifier it relies on, reaching 65.0% accuracy against 71.7% on a balanced 60-case sample, despite having access to model explanations, graph context, and retrieved reference cases. Of the eight decisions the agent changed, six replaced correct classifier outputs with errors, and it produced a coherent written rationale in each case. An exploratory disagreement-based escalation rule flagged two of these agent errors for human review without flagging any correct decision. We conclude that each component of a layered fraud system contributes only under specific conditions, and that a plausible rationale from an investigation agent is not evidence of a better decision.
The typical approach to fraud detection in retail environments relies on rules-based, static detection engines. Such systems perform well against simple, known attack types but struggle against sophisticated, emerging threats and organized financial syndicates. Moreover, rules-based systems produce high false positive rates that seriously degrade the experience of legitimate customers. This paper presents a Graph Artificial Intelligence framework to support high-throughput and streaming retail transactions with built-in explainability. Our approach treats transactions as a dynamic and heterogeneous graph to capture complex relational dependencies and structural anomalies in real time. We validate the framework using a curated subset of the publicly available IEEE-CIS Fraud Detection dataset under a streaming scenario, evaluating both latency and predictive accuracy. Combined with PyTorch Geometric for graph neural network construction and SHAP (Shapley Additive exPlanations) for post-hoc model interpretation, our system delivers high prediction accuracy and human-interpretable decision making. The proposed model shows strong accuracy and recall compared to traditional baselines, while flagged transactions are explained through visual subgraph evidence. This transparency enables fraud analysts to validate alerts efficiently, reducing review time and operational cost in e-commerce environments.
Jose Prabhu Michael Singarayan, Jayakumar Ramalingam, Neetu Uthaman· 2026 International Conferenc...· 0 citations
Financial statement fraud inflicts large and recurring losses on capital markets, yet the dominant detection paradigm still relies on single, black-box classifiers (e.g., RUSBoost) trained on structured accounting ratios alone. Two limitations follow: (i) the rich, unstructured Management Discussion and Analysis (MD&A) narrative of the 10-K filing is discarded, and (ii) the resulting scores are difficult for auditors to trust because they carry no transparent, standards-aligned rationale. Recent large language model (LLM) systems have shown that multi-agent collaboration is more robust than a single LLM for anomaly detection, but no study has systematically transferred this paradigm to listed-company statement fraud. We propose FraudDebate-Agent, a four-role multi-agent system in which a Quantitative Analyst agent scores 28 raw accounting items and 14 ratios with gradient-boosted and tabular attention models, a Narrative Auditor agent quantifies tone, linguistic uncertainty, and year-over-year textual novelty of the MD&A with FinBERT, and an Industry Peer agent uses retrieval-augmented generation to measure industry-relative anomaly. A Critic–Debate agent then orchestrates a pair-wise Evidence-based Multi-Agent Debate (EMAD) that reconciles disagreement across modalities and arbitrates a reconciled fraud-risk assessment, which is aggregated over a tri-modal evidence graph. Our contributions are as follows: (1) the first use of an evidence-grounded debate mechanism for accounting fraud, which materially reduces LLM hallucination; (2) a numerical–textual–peer evidence graph that fuses heterogeneous signals; and (3) an explainable report aligned with the PCAOB AS 2401 fraud-risk taxonomy. On AAER-labelled firm-years linked across a SEC financial dataset and EDGAR-CORPUS, FraudDebate-Agent improves the area under the ROC curve and the rare-event ranking metric NDCG@k over the strongest single-modality and single-LLM baselines while producing substantially more faithful explanations. We frame the system as a fraud-risk screening and risk-ranking tool for AAER-labelled misstatement risk rather than a determination of fraudulent intent. We report results over multiple seeds to reflect real-world stochasticity and discuss limitations and cross-domain applications.
Sophisticated machine learning and deep learning architectures often achieve high predictive performance but remain insufficient for high-stakes domains such as financial fraud detection, where stakeholders require both accuracy and trust in the decision-making process. This study proposes a weighted ensemble framework that strategically combines eXtreme Gradient Boosting (XGBoost) and Graph SAmple and aggreGatE (GraphSAGE) models to optimize the precisionrecall trade-off while maintaining interpretability. Our proposed framework, BoostSAGE, leverages complementary strengths of its constituent models. XGBoost provides high precision (94%) through transaction-level feature analysis, while GraphSAGE ensures superior recall (95%) by capturing structural risk signals across transaction-user-merchant relationships. Empirical results indicate that BoostSAGE outperforms constituent models by synthesizing local and relational feature sets. Most notably, the ensemble delivers a 14.97% improvement in AUCPR over the XGBoost baseline. By leveraging historical fraud aggregation within user-centric connections, the model reduces total False Negatives by 44.8% compared to XGBoost, yielding a final F1score of 92.88%. Our results demonstrate that weighted ensemble framework provides a robust and interpretable approach to fraud detection, balancing operational efficiency with stakeholder trust.
Chaw Thet Zan, Yusuke Maruyama, Felix Beierle· Annual International Compute...· 0 citations
Detecting fake-order fraud at scale remains a critical challenge for large online-to-offline (O2O) service platforms, as existing approaches often rely on expert-designed features, produce black-box decisions, and provide limited interpretability. To address these limitations, we propose DeepScrub, a reinforcement learning framework built upon large language models (LLMs) for fake-order fraud detection with traceable reasoning. DeepScrub introduces three innovations. First, a semantic unification module converts heterogeneous risk signals into textual descriptions that LLMs can understand. Second, continued pre-training on risk-control corpora injects domain knowledge, and task rewards jointly evaluate prediction correctness and reasoning quality. Third, the SUggest-REflect (SURE) mechanism incorporates expert feedback and model self-checking to iteratively refine reasoning paths. On a real-world fake-order fraud detection dataset, DeepScrub achieves a macro-F1 score of 85.3%, outperforming the best baseline by 2.7 percentage points. Our task-optimized 8B model further surpasses a 32B model, showing that domain adaptation can matter more than model scale in this setting. In a four-week live pilot, DeepScrub achieved 91.8% precision and 88.5% recall, improving over first-stage human reviewers by 16.6 and 38.8 percentage points. It reduced first-stage manual review workload by 94% and saved nearly one million RMB annually. These results show that DeepScrub improves fraud review accuracy, reduces first-stage review workload, and provides traceable evidence for production risk-review workflows.
Siqi You, Bingsong Xu, Zhixiang Zheng et al.· 0 citations
Credit card fraud is a menace to financial institutions, but detection is compromised by highly imbalanced transaction datasets. This study proposes an advanced machine learning framework optimized for fraud detection. To address the issue of data imbalance, SMOTE-Tomek Links is applied to synthetically generate minority fraud cases while removing noisy, overlapping majority-class instances. Recursive Feature Elimination (RFE) is deployed to identify the optimal features, and RandomizedSearchCV automates hyperparameter optimization. The study introduces a Stacked Logistic Regression ensemble to combine the predictive capacity of optimized Random Forest and XGBoost base classifiers. The model’s effectiveness is assessed using seven evaluation methods: accuracy, recall, precision, confusion matrix, F1-score, Receiver Operating Characteristic Area Under the Curve (ROC-AUC) score and the Area Under the Precision-Recall Curve (AUC-PR) score. Findings reveal that the proposed stacked model performance surpasses both individual base models. While achieving deceptively high baseline accuracy across all models, the stacked ensemble delivers a superior AUC-PR score of 0.8207 and an F1-score of 0.93. This minimizes the confusion matrix misclassifications to just 20 False Negatives and 5 False Positives. The framework provides a cost-optimized operational engine that aggressively mitigates bank fraud losses while successfully shielding legitimate cardholders from accidental checkout declines.
Uduh Israel Akakoh, G. N. Edegbe· FUDMA Journal of Sciences· 0 citations
An Explainable Artificial Intelligence (XAI) framework tailored for banking transaction anomaly detection within internal audit workflows is introduced, confirming that feature-level explanations measurably improve auditor confidence and decision quality.