Adversarial fraud sample generation with reinforcement learning: A joint optimization framework for financial statement fraud detection

The emergence of the Unified Payments Interface (UPI) has transformed the digital payment ecosystem by enabling seamless, instant, and interoperable financial transactions across India. Its widespread acceptance has accelerated the shift toward cashless payments and increased the availability of digital financial services for millions of users. However, the rapid growth in transaction volume and user adoption has also created new opportunities for cybercriminals to exploit vulnerabilities within the digital payment infrastructure.Financial fraud associated with UPI platforms has become increasingly sophisticated, involving techniques such as phishing campaigns, fraudulent QR codes, identity impersonation, account hijacking, social engineering attacks, and the misuse of mule accounts. These evolving threats generate complex transaction patterns that are often difficult to detect using conventional rule-based security mechanisms. As fraud strategies continue to change, static detection systems struggle to provide accurate and timely identification of suspicious activities.Recent progress in Artificial Intelligence, Machine Learning, and Deep Learning technologies has significantly enhanced the capability of fraud detection systems. One of these developments has caught a lot of interest from researchers is hybrid deep learning methods that are able to integrate the benefits of several computational models. Several techniques can be combined in a unified approach for enhancing fraud detection, such as Convolutional Neural Networks (CNNs), Long Short-Term Memory (LSTM) networks, Autoencoders, Graph Neural Networks (GNNs), Explainable Artificial Intelligence (XAI), and Federated Learning. These architectures enable real-time analysis, adaptive learning, privacy protection, and efficient management of large-scale transactional data. This review is based on the recent research and studies regarding the detection of fraudulent activities in UPI-based payment environment using hybrid deep learning techniques. It offers detailed evaluations of the current models, their strengths and weaknesses, explores current issues and challenges, and suggests areas of future work. These results suggest that hybrid deep learning architectures are promising to improve the security of digital payment systems through better detection accuracy, fewer false alarms, safeguarding sensitive user data, and greater transparency of automated decision-making. Thus, these wise frameworks are a valuable basis for developing safe, reliable, and trustworthy digital financial ecosystems

As trust, auditability, and the long-term security of cryptographic mechanisms become increasingly critical, modern digital payment systems are adopting intelligent decision-making techniques for fraud detection under stringent latency and reliability constraints. However, existing fraud detection approaches primarily focus on improving classification accuracy within isolated systems, relying on traditional security rules, basic cryptographic schemes, and centralized authentication mechanisms. Such approaches are insufficient to address evolving fraud patterns, holistic system trust requirements, and emerging post-quantum threats. To address these limitations, this work proposes an end-to-end, risk-adaptive transaction security framework that reconceptualizes fraud detection as a dynamic and continuous security process rather than a stand-alone prediction task. The framework introduces Hybrid Anomaly-Guided Adaptive Detection (HAGAD), which integrates supervised fraud prediction, unsupervised anomaly detection, and temporal behavior analysis to generate a continuous risk score. This risk score dynamically governs Adaptive Key Ratcheting (AKR), risk-triggered association quorums, and fault-tolerant post-quantum threshold signing, while selectively employing a permission blockchain for high-risk transactions. Furthermore, the system incorporates an Adaptive Assurance Scoring (AAS) mechanism to evaluate trustworthiness beyond conventional accuracy-based metrics. Experimental results demonstrate that the proposed approach achieves high reliability with balanced precision and recall, while activating enhanced security mechanisms only when necessitated by transaction risk. This adaptive enforcement ensures robust decision validation without introducing additional overhead for low-risk transactions. Overall, the proposed framework establishes an integrated, risk-adaptive transaction security pipeline that unifies intelligent detection, advanced cryptography, auditability, and assurance evaluation, providing a resilient and future-ready foundation for next-generation electronic payment systems.

Recent regulation, the EU AI Act and ISO/IEC42001, requires organizations deploying high-risk artificial intelligence (AI) systems to identify, assess, and document the risks those systems generate, including risks that arise from the opacity of the underlying models. Existing risk-assessment methods leave a gap between checklist-style failure mode and effects analysis (FMEA), which is too informal for the new regulatory regime, and full probabilistic risk assessment, which is impractical for opaque deep-learning systems whose component-level failure data are not available. This paper presents a hybrid framework that adapts FMEA to AI failure modes spanning data, model, infrastructure, and ethical dimensions, and uses the analytic hierarchy process (AHP) to derive domain-specific criterion weights with an explicit consistency check. A fourth criterion, organizational impact, is added to the conventional severity, occurrence, and detectability triplet to capture the systemic regulatory and reputational consequences that traditional FMEA does not represent. The framework is illustrated on three case studies in healthcare imaging, financial fraud detection, and autonomous vehicle perception, with cross-domain weights elicited under acceptable consistency. A Monte Carlo sensitivity analysis and a cross-method robustness check against TOPSIS, ELECTREIII, and PROMETHEEII indicate that the resulting prioritization is stable both under weight perturbation and under change of aggregation logic. The framework is presented as an auditable governance tool for pre-deployment risk review, not as a replacement for FMEA in domains where it already works, and validation with independent expert panels is identified as the critical nextstep.

Voice phishing represents a rapidly evolving form of cyber-enabled financial fraud that exploits telecommunications networks to deceive victims. Traditional reactive policing approaches face significant limitations in preventing such crimes in real time. This study introduces a data-driven framework for the proactive detection of phone numbers linked to voice phishing operations using large-scale call log data from South Korea. Behavioral features derived from call metadata were utilized to capture distinctive communication patterns that differentiate fraudulent users from legitimate ones. A stepwise logistic regression model was initially employed to identify key predictors of fraudulent behavior, followed by advanced machine learning models-including random forest, gradient boosting, and Multi-Layer Perceptron (MLP)-for classification. The results reveal that voice phishing numbers display unique behavioral characteristics, such as concentrated weekday activity, shorter call durations, higher outgoing call ratios, and a preference for Mobile Virtual Network Operator (MVNO) usage. The proposed models achieved high performance, exceeding 95% accuracy and 97% recall, demonstrating their robustness in detecting suspicious numbers. These findings underscore the potential of integrating artificial intelligence and behavioral analytics into proactive fraud detection systems and contribute to the development of early-warning mechanisms for preventing telecommunication-based financial crimes.

Detecting collusion rings in modern banking requires modeling the evolving structural interactions among heterogeneous entities (customers, accounts, and devices) rather than isolated transaction features. Most graph-based fraud detectors assume abundant labels, yet confirmed fraud labels in anti-money laundering (AML) settings routinely arrive months after the fact. We introduce Audit-HCL, a dynamic heterogeneous graph neural network framework that uses dual-view contrastive learning to operate effectively under this label scarcity. Audit-HCL represents the transaction ecosystem as a temporal sequence of heterogeneous graph snapshots, encodes them through a metapath-guided heterogeneous attention encoder, and tracks evolving node behavior with a GRU-based temporal dynamics module. A cross-view contrastive objective aligns structural and temporal perspectives for legitimate nodes while separating anomalous ones, guided by an anomaly-aware negative sampling strategy. Experiments on two public benchmarks (Elliptic and IBM AML-Synthetic) show that Audit-HCL outperforms fourteen baselines by 3.2% in AUC-ROC and 6.8% in F1-score, with the gains over the strongest competitors confirmed by paired significance tests, and that it retains useful discriminative power with zero fraud labels. On the synthetic IBM AML benchmark, it also detects laundering patterns an average of 7.4 weeks ahead of confirmed events ([Formula: see text] the lead time of the best baseline) by capturing gradual structural drift before large-scale fund transfers begin, although the magnitude of this lead time is tied to the controlled typologies of the synthetic data and should be read as indicative rather than as a guarantee for production AML environments.

Many organisations collect sensitive data that cannot be freely shared. Hospitals store brain magnetic resonance imaging (MRI) scans on internal servers; banks keep transaction records behind strict firewalls; agricultural services retain crop images in isolated repositories. Federated learning (FL) allows models to be trained without centralising raw data, yet most existing systems address a single domain and offer limited insight into model behaviour and provenance over time. BlockFedX is a cross-domain federated learning system designed to address three simultaneous tasks: credit card fraud detection on tabular data, brain tumour detection on MRI images, and plant disease recognition on leaf images. These three domains were deliberately selected because they represent the principal data modalities in real-world privacy-sensitive deployments-structured tabular records, greyscale medical images, and colour natural images-and because public benchmark datasets exist for all three, enabling reproducible evaluation. The system uses a shared backbone that is updated only where model layers have compatible tensor shapes, while domain-specific output layers remain local at each client. Explanations are computed at the clients using SHAP feature-attribution for tabular data and Grad-CAM visual heatmaps for images; the server receives only compact statistical summaries. The server also applies a distance-based anomaly test on client updates and records model hashes, explanation summaries, and anomaly flags in a hash-chained ledger. Experiments on three public datasets under non-identical client data distributions show that BlockFedX achieves an average fraud-detection F1-score of 0.92, 74.32% mean validation accuracy on BrainMRI, and 77% test accuracy on PlantVillage, while keeping all raw data local. These results are below strong centralised baselines, as expected under compact models and non-IID splits, but the system simultaneously provides three properties rarely combined in prior work: cross-domain federated training via a shape-safe backbone, client-side explanations integrated into the learning loop, and a lightweight tamper-evident record of model evolution across rounds.

Public pension systems in the United States disburse trillions of dollars in retirement and disability benefits each year. A persistent and growing challenge for these systems is the occurrence of improper payments, which encompasses both administrative errors and deliberate fraudulent acts. The U.S. Government Accountability Office (GAO) reported $162 billion in improper federal payments for fiscal year 2024, of which approximately 84 percent were overpayments. The Office of Personnel Management (OPM) alone improperly paid $245 million in overpayments during fiscal year 2022, with unreported annuitant deaths identified as the leading cause. This article proposes and evaluates a multi-layered data analytics framework designed specifically for detecting overpayment and fraud in public pension systems. The framework integrates supervised machine learning classifiers, logistic regression, random forest and gradient boosting with unsupervised anomaly detection methods, including isolation forest and local outlier factors. The study used a simulated federal pension dataset, which was constructed from publicly documented fraud typologies. The proposed framework achieves a precision of 93.4 percent and a recall of 91.7 percent under the random forest classifier after SMOTE-based class balancing. The results are compared against baseline rule-based detection systems currently employed by federal agencies. The article further discusses policy implications for the Payment Integrity Information Act of 2019, the Social Security Administration Death Master File data-sharing program and the Treasury's Do Not Pay initiative. This research advances literature by applying predictive analytics to a domain that has received comparatively little attention relative to the severity of its fiscal impact. Keywords: Pension Fraud Detection, Overpayment Analytics, Machine Learning, Anomaly Detection, Public Sector Integrity, Improper Payments, Random Forest, Federal Benefits.

Credit card fraud detection is a difficult applied machine learning problem. It combines extreme class imbalance, temporal non-stationarity, and a sharp cost gap between missed fraud and false alarms. This paper presents an end-to-end experimental framework for fraud detection on the benchmark European transaction dataset (284,807 transactions; fraud prevalence 0.173%). A strict no-data-leakage protocol is enforced throughout. The data are split chronologically into training (70%), validation (15%), and test (15%) sets, and every preprocessing step - feature scaling, SHapley Additive exPlanations (SHAP)-based feature selection, and oversampling - is fitted only on the training partition. Two domain-informed features are engineered from the raw timestamp (sinusoidal hour encoding and log-transformed amount), and SHAP analysis reduces the 33-dimensional feature space to 15 features. Six oversampling strategies - SMOTE, BorderlineSMOTE, SVMSMOTE, ADASYN, SMOTEENN, and SMOTETomek - are compared across 12 classical classifiers, 3 multilayer perceptron (MLP) architectures, a purpose-built deep neural network (FraudNet), and 7 ensemble methods, giving 85 model-sampler combinations. Decision thresholds are tuned on the validation set using the [Formula: see text]-score, and all final metrics are reported on the held-out temporal test set. To characterise temporal drift explicitly, we measure distributional shift between splits using the Population Stability Index (PSI), Kolmogorov-Smirnov (KS) tests, and Jensen-Shannon divergence on the SHAP-selected features. We also report 1000-replicate bootstrap 95% confidence intervals for the leading configurations. The MLP (128-64-32) without oversampling reaches the highest individual [Formula: see text] of 0.7722 (95% CI: [0.6712, 0.8420]). The Soft Voting ensemble attains the best Matthews Correlation Coefficient (MCC) of 0.8060 (95% CI: [0.7045, 0.8807]) and an AUC of 0.9703. LightGBM under SMOTEENN shows the largest gain from oversampling, with [Formula: see text] rising from 0.0745 to 0.7588. ADASYN consistently underperforms, and no single oversampling method dominates across all model families. The drift analysis confirms measurable but modest distributional shift between splits. Because the dataset spans only 48 hours, this shift reflects short-horizon, mostly intra-day variation rather than the long-horizon concept drift seen in production; we therefore treat the chronological protocol as a methodological lower bound and emphasise this limit throughout. Together, these findings give practical guidance for designing production-grade fraud detection systems under strict temporal and data-integrity constraints.

BackgroundMedicaid administrative spending funds core program operations such as fraud and waste detection and prevention, technological infrastructure, and eligibility determination. Though typically lower than private insurance administrative costs, such spending varies across states and may respond to policy and economic shocks.PurposeThis study examined trends in Medicaid administrative expenditures before and during the federal maintenance of effort (MOE) policy enacted during the COVID-19 Public Health Emergency (PHE).MethodsUsing state-level panel data from 2018–2022 for all 50 states and the District of Columbia, we regressed Medicaid administrative spending outcomes (percent changes in dollar amount, per-enrollee, and per capita spending) on a binary indicator for MOE period and controlled for unemployment rates, enrollment changes, and managed care characteristics using linear regression models with state fixed effects. We also estimated models stratified by administrative burden tertiles.ResultsPercent change in Medicaid administrative spending (amount in United States dollars) remained stable during the MOE period relative to pre-MOE years. Per-enrollee administrative spending declined by 9.1 percentage points.ConclusionsDespite record enrollment under MOE policies, Medicaid administrative spending did not significantly increase. The findings suggest that states maintained administrative cost stability, potentially achieving economies of scale, and that administrative burden levels and other time varying state characteristics did not meaningfully influence these trends.

ABSTRACT The rapid development of AI, big data, and cloud computing technologies has significantly impacted the function of the enterprise audit system, which has transformed from an intelligent management system that performs intelligent analysis, intelligent prediction, and intelligent early warning of problems based on large amounts of data to a complex digitalized business environment in which much attention has been given to the function of a good fraud detection system. In this study, we explored the application value of audit analytics AI capability within the context of an enterprise auditing system. A structured research-based questionnaire was sent out to hundreds of internal auditors, IT auditors, compliance managers, and related professionals. The analysis was performed using a combination of descriptive statistics, regression analysis, and SEM techniques. The results from our analysis were substantial in demonstrating the potential of AI-powered audit analytics in improving the accuracy of fraud detection, reducing the time to respond to incidents, and identifying potential risks. The research contributes to the development of both intelligent auditing and enterprise governance, and it offers practical recommendations and solutions for organizations to prevent and identify fraudulent activities and to improve their auditing efficiency through incorporating AI-powered enterprise auditing systems.

Machine learning has become central to financial fraud detection, yet a growing body of evidence indicates that the way such models are evaluated in the academic literature diverges sharply from the conditions under which they are deployed. Reported gains are frequently driven by methodological artefacts, such as random train and test splitting of temporally ordered transactions, resampling applied before splitting, and reliance on accuracy or area under the receiver operating characteristic curve, rather than by genuine improvements in fraud-catching capability. This article argues that the field suffers from a benchmark-to-deployment gap and proposes a unified evaluation protocol intended to close it. The protocol rests on four principles: strictly temporal data partitioning that forbids the use of future information, simulation of realistic label delay to reflect the months-long interval before fraud is confirmed, cost-sensitive and operationally meaningful metrics aligned with the economics of false positives and false negatives, and full reporting of the conditions under which results are obtained. The protocol is positioned against the datasets and practices that dominate current research, and a re-evaluation methodology is described whereby previously published models are reassessed under realistic conditions. It is shown that conventional evaluation can inflate apparent performance substantially and can reorder the relative ranking of methods, so that models presented as state of the art under random splitting may underperform simpler baselines once temporal integrity and cost are respected. The article concludes that adopting a shared, deployment-aware evaluation protocol is a precondition for meaningful scientific progress in fraud detection and offers a concrete, reproducible specification to that end.

ABSTRACT Despite the maritime supply chain being the backbone of global trade, it faces persistent challenges in transparency, fraud prevention, shipment tracking and data privacy. Blockchain technology has emerged as a transformative solution, enhancing trust and traceability within supply chain networks. However, its limitations in data privacy and scalability necessitate advanced privacy-preserving mechanisms. Zero-Knowledge Proofs (ZKP) offers a cryptographic approach to validate data without exposing sensitive information, addressing blockchain’s privacy constraints. This paper reviews the state of the art on current applications of blockchain in maritime supply chain management and explores the integration of ZKP for secure trade document verification, fraud detection, privacy-preserving traceability and regulatory compliance. Additionally, it examines computational overhead, scalability and adoption barriers while proposing future research directions. Implementing ZKP within blockchain-based port operations enables robust governance models, ensuring data verification without revealing confidential details. This approach fosters a secure and privacy-compliant trade environment, enhancing trust and collaboration among stakeholders. By optimising resource allocation and mitigating risks, integrating ZKP can significantly improve maritime supply chain efficiency. Integrating Zero-Knowledge Proofs with blockchain, maritime logistics can achieve a balance between transparency, security and operational efficiency, addressing existing challenges in data privacy and regulatory compliance, improving the sustainability of port operations.

ABSTRACT The rise of e-commerce in India has revolutionized the way consumers shop, with sites like Amazon, Flipkart and Myntra handling millions of transactions each day. That expansion has, however, led to commensurate cybersecurity dangers such as phishing, hacking, ID theft, payment fraud, data breaches, and more. This study examines the implications of cybersecurity threats on e-commerce operations and the consumers’ trust in the e-commerce in India and highlights the major vulnerabilities and analyzes the effectiveness of cybersecurity countermeasures like encryption, multi-factor authentication (MFA), and AI-driven fraud detection. Primary survey data is collected from 150 respondents, and secondary analysis of published industry and regulatory reports is used to create a mixed methods approach. Research shows that 39 percent of consumers surveyed feel unsafe while making online payments, 34 percent have been victims of cyber fraud, and 48 percent of consumers are aware of common cyber frauds like phishing. The study shows that strong cybersecurity governance is not just a technical requirement but also a strategic lever to achieve sustainable growth in e-commerce in India. Suggestions are provided for IT auditors, information security professionals, and e-commerce platform operators.

Financial fraud detection requires screening massive transaction networks where evolving topologies, extreme label sparsity, and asymmetric misclassification costs make traditional classification paradigms ineffective. We propose ST-CGNN, a spatio-temporal contrastive graph neural network that frames operational screening as a multi-task learning problem in which a shared encoder is supervised by a contrastive regularizer and an evidential triage head. Concretely, ST-CGNN combines a continuous-time heterogeneous encoder with a hard-negative contrastive regularizer and an evidential output head, so that structural representations and uncertainty-aware prioritization are trained from a common backbone with summed losses rather than as a sequential, modular pipeline. Evaluated under strict chronological constraints on large-scale public and controlled benchmarks, ST-CGNN consistently outperforms state-of-the-art GNNs and post-hoc calibration methods. Specifically, on the DGraph-Fin benchmark, the proposed evidential triage score improves Precision@100 to 0.884 and achieves a calibrated ECE of 0.034. On Elliptic, the difference between ST-CGNN and the best competitor (MTP-GAT) lies within seed variance and is not statistically distinguishable; gains concentrate on benchmarks where heterogeneity and bursty timing dominate. Paired bootstrap tests and selective-prediction analysis confirm that this shared-encoder design significantly enhances the reliability of fixed-budget analyst reviews, providing a robust foundation for high-stakes risk management in dynamic transaction environments.

Research note: A machine learning approach for authentication of laying hen housing systems based on egg quality parameters.

Comprehensive metabolic profiling of 82 unifloral Jamun honey samples from the Mahabaleshwar region (Apis cerana) was achieved using NMR spectroscopy, HPLC, solvent extraction, and targeted biophysical assays. This is by far one of the largest datasets characterised for a single type of honey. • Exceptionally low glucose and low total glucose+fructose content observed for the Jamun honey are among the lowest reported for honeys in general. These features are consistently observed across all 82 samples, yielding a high fructose-to-glucose ratio , indicative of a potentially low glycemic index . • NMR signatures showed minimal fermentation markers across all samples, demonstrating the intrinsic stability of Jamun honey despite its tropical origin. • HPLC analysis identified six phenolic acids and six flavonoids , including high levels of myricetin, kaempferol, and 4-hydroxybenzoic acid , underpinning the honey’s strong antioxidant capacity, confirmed through TPC, TFC, and antioxidant assays. • Melissopalynology revealed sparse Jamun pollen , emphasizing its limited reliability for floral authentication and demonstrating the need for metabolomics-driven approaches. • The study delivers the first robust reference dataset for saccharides, metabolites, and antioxidant markers in authentic Jamun honey, establishing a scientific foundation for quality control, fraud detection, and honey authentication . • The metabolic profile positions Jamun honey as a high-value medicinal honey , comparable to Manuka, with elevated antioxidant metabolites and relatively lower sugar levels. Jamun honey, produced by Apis cerana bees foraging on Syzygium cumini (Malabar plum), is traditionally recommended for diabetics and valued for its antioxidant and medicinal properties. This study analyzes 82 Jamun honey samples from Mahabaleshwar, India for a detailed metabolite profiling using NMR spectroscopy and HPLC. NMR spectroscopy was employed to profile the sugar composition of honey, enabling quantitative analysis of glucose, fructose, maltose, sucrose, and turanose. Polyphenols and flavonoids were isolated by solid-phase extraction and subsequently identified and quantified using HPLC. The antioxidant properties of the Jamun honey were further characterized through biophysical assays. NMR revealed generally lower glucose (25.91±0.28 g/100 g of honey) and lower total glucose+fructose levels (61.22±0.64 g/100 g of honey)- compared to other honeys- resulting in a high fructose-to-glucose ratio (1.36) and suggesting a potentially low glycemic index. All samples showed minimal fermentation markers (hmf at 0.0003±0.0007 g/100 g of honey, ethanol at 0.018±0.0006 g/100 g of honey), indicating excellent stability. HPLC and antioxidant assays confirmed high antioxidant capacity (85.10±3.50 %), with elevated levels of myricetin, kaempferol, 4-hydroxybenzoic acid, and nine other polyphenols. Pollen analysis showed low Jamun pollen content, underscoring limitations of melissopalynology for floral authentication. In the context of rising honey adulteration and honey quality control, this study provides a compositional reference for authentic Jamun honey using advanced NMR and data-based profiling approaches on a large sample set.

Dual-range colorimetric protein quantification in whey and WPC using a smart polymeric film and RGB imaging.

• Wheat flour fraud risk is rising during global disruptions like the war in Ukraine. • Portable Raman and IR detected wheat flour adulteration with potato and corn flour. • IR had a 6.9% detection limit, while Raman showed a higher limit of 22.6%. • Portable vibrational spectroscopy offers rapid, non-destructive detection of flour fraud. Monitoring wheat flour fraud is vital, especially during global disruptions like the war in Ukraine, which heighten the risk of economically motivated adulteration. This study explores the use of portable Raman and infrared (IR) spectroscopy to detect wheat flour adulteration with potato and corn flour. Principal Component Analysis (PCA) revealed spectral differences, while Partial Least Squares Regression (PLSR) models quantified adulterant levels. Both techniques successfully identified adulteration, with typical errors of prediction for test samples of 4-5% for infrared and 8.7-10.6% for Raman spectroscopy. In addition, IR achieved a limit of detection (LOD) as low as 6.9% and Raman 22.6%. The findings highlight portable vibrational spectroscopy as a rapid, non-destructive tool for detecting flour fraud, supporting food quality control and regulatory enforcement

SRAF: A structure-aware relation-guided model for fraud detection on heterogeneous graphs