This work constructs and makes publicly available a comprehensive U.S. company dataset combining financial statements, summarized MD&A text, and fraud labels and achieves the best performance on the challenging CI-FSFD task, demonstrating the critical value of textual data and robust evaluation for reliable financial fraud detection.
Abstract
Financial statement fraud detection (FSFD) is crucial for market integrity but faces challenges from increasingly sophisticated schemes and under-utilized textual data in financial reports. Existing methods often rely on random data splits, leading to overoptimistic performance estimates that do not reflect real-world generalization to new companies or future periods. To address this recurring problem with the state of the art, we propose a robust FSFD framework leveraging Large Language Models (LLMs) to integrate both structured financial data and unstructured textual information from financial reports. We provide a more realistic evaluation through a novel and challenging benchmark task called Company-Isolated FSFD (CI-FSFD). We construct and make publicly available a comprehensive U.S. company dataset combining financial statements, summarized MD&A text, and fraud labels. Our approach achieves the best performance on the challenging CI-FSFD task, demonstrating the critical value of textual data and robust evaluation for reliable financial fraud detection.
The increasing complexity of digital financial systems has reshaped financial fraud detection from isolated transaction classification into relational risk reasoning over interconnected financial entities. This shift has motivated graph-based fraud detection, where models identify fraudulent nodes by exploiting dependencies among customers, cards, merchants, categories, and locations. However, despite rapid progress in graph-based methods, existing public benchmarks remain misaligned with real-world financial systems in two important aspects. First, they often simplify financial ecosystems into homogeneous or single-node-type multi-relational graphs, failing to preserve the multi-entity and multi-relational nature of financial data. Second, they rarely provide large-scale heterogeneous financial graph datasets with realistic operating conditions such as extreme class imbalance and limited label availability, making it difficult to assess the practical effectiveness of current methods. To address these gaps, we present FinFraudBench, a heterogeneous graph benchmark for financial fraud detection. FinFraudBench contains two heterogeneous graph datasets (CreditCard-Fraud and BankTrans-Fraud) with up to 8.99M nodes and 89.23M directed typed edges. Each dataset preserves six financial entity types, fourteen directed edge types, and natural fraud rates that mirror deployment constraints. With these datasets, we establish a standardized evaluation protocol covering both ranking and imbalance-sensitive classification metrics, and evaluate representative baselines. Extensive experiments yield empirical insights into current methods'limitations and suggest promising avenues for future research. FinFraudBench is available at https://anonymous.4open.science/r/FinFraudBench-B002.
Yixuan Chen, Hongyu Zhan, Jie Sheng et al.· 0 citations
Detecting financial statement fraud from narrative disclosures is difficult because evidence is sparse and distributed across long, structured public-company annual reports filed on Form 10-K. We present a time-aware benchmark focused on Management’s Discussion and Analysis (MD&A) for financial statement fraud detection (FSFD) and evaluate representative long-document architectures under time-aware forward-chaining splits. We compare hierarchical encoders (hierarchical attention transformer and hierarchical document transformer), a finance-domain long encoder (LongFinBERT), a sparse-attention model (Longformer/LED), and a short-context head+tail BERT baseline. Performance is assessed using the area under the receiver operating characteristic curve (AUROC), average precision, audit-budget ranking quality based on cumulative gain, and deployability metrics such as peak GPU memory and per-filing latency. We further propose an evidence-driven Hybrid pipeline (Selector → Reranker → Calibrated Stacker) that selects paragraph evidence with high recall, applies cross-evidence reranking, and fuses calibrated scores with a monotone stacker. On the held-out 2014–2019 horizon, the Hybrid achieves AUROC = 0.871 and average precision = 0.392 and improves audit-budget ranking over the strongest single model. In our benchmark, hierarchical encoders offer the strongest single-model accuracy–efficiency trade-off, outperforming the sparse-attention baseline at lower memory and latency, while head+tail BERT remains a competitive low-cost anchor. Year- and issuer-stratified block bootstrap analysis indicates that gains in average precision and cumulative-gain ranking quality are robust under our time-aware evaluation protocol.
Matin N. Ashtiani, B. Raahemi· IEEE Access· 0 citations
Cheque fraud is a material risk in after-hours business deposit operations because funds may be released within one business day, while cheque clearing takes several days. This timing gap creates a fraud exposure window for financial institutions. Prior mitigation relies on static, deposit-level checks and therefore miss historical client behavior and evolving patterns. To address this gap, we propose a multi-view ensemble ML framework that combines: Extreme Gradient Boosting (XGBoost) for known fraud patterns, Isolation Forest for label-free anomaly detection, and Graph Sample and Aggregate (GraphSAGE) for relational patterns associated with transaction activities. We then combine the three outputs into a single client-level risk score. Under stable conditions, performance is comparable to XGBoost; under a targeted distribution shift, our framework performs best (F1: 83.77%, FPR: 0.69%) versus XGBoost (F1: 82.77%, FPR: 0.72%). These results indicate improved robustness to distribution shift while preserving interpretability through plain-language explanations grounded in behavioural, anomaly, and relational evidence.
Ann Youduo Xu, Emily Yu, Justin Leski et al.· 0 citations
Fraud detection in digital financial transactions is traditionally formulated as a binary classification problem, although real-world fraud investigation requires analysts to prioritize a limited number of suspicious transactions according to operational risk and potential financial impact. This study reformulates fraud detection as an amount-aware operational ranking problem for fraud-risk prioritization. Transactions are organized into time-window query groups, and fraudulent transactions are assigned graded relevance based on training-only transaction-amount quartiles, enabling the ranking objective to distinguish low- and high-severity fraud without relying on proprietary cost matrices. The proposed formulation is implemented using a representative Learning-to-Rank framework based on LambdaMART, while an out-of-fold XGBoost risk score is incorporated as an auxiliary feature to refine the ranking representation rather than serve as the primary contribution. Experiments conducted on a public credit-card fraud dataset using chronological validation and future-holdout testing demonstrate that amount-aware relevance consistently improves severity-aware top-rank ordering compared with conventional binary relevance. The proposed HybridLTR_amount model significantly outperforms XGBClassifier and PureLTR_binary in terms of all-query NDCG@10, whereas its performance is not statistically different from PureLTR_amount, indicating that the primary empirical improvement is attributable to the amount-aware ranking formulation rather than the auxiliary hybrid component. Additional operational analyses show that high-risk transactions and fraudulent financial losses are concentrated within a compact top-ranked segment, while budget-oriented evaluation demonstrates the practical value of the proposed formulation under limited analyst review capacity. These findings establish amount-aware operational ranking as an effective formulation-centric framework for operational fraud-risk prioritization rather than as a new classification algorithm.
H. Hartatik· Journal of Computing Theorie...· 0 citations
This research aimed to develop and evaluate a financial statement fraud (FSF) detection model for companies listed on the Indonesia Stock Exchange (IDX) during the 2024–2025 period using a machine learning approach. An ablation study design was employed to test 27 model combinations across three feature scenarios: Beneish M-Score financial ratios, linguistic features extracted from Management Discussion and Analysis (MD&A) texts using the InSet lexicon, and 30-day stock price volatility. Nine classification algorithms were evaluated using precision, recall, and F1-score metrics. The best-performing model combined financial ratios and linguistic features using Gradient Boosting, achieving an F1-score of 0.545. In contrast, the addition of stock price volatility as a feature did not improve model performance and instead reduced the classification ability of all tested algorithms. These findings indicate that the Indonesian capital market may have limited ability to anticipate indications of financial statement fraud before related information is publicly disclosed. This study concludes that integrating accounting and linguistic information is more effective than relying solely on financial ratios or incorporating market data. These findings contribute to the development of machine learning-based FSF detection literature in Indonesia and provide an alternative approach for auditors, investors, and regulators to enhance the effectiveness of early financial statement fraud detection.
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
