Aug 2026· International Scientific Journal of Engineering and Management· Vol 05, pp. 1-9· 0 citations
TL;DR
A conceptual, layered fraud-detection framework is proposed that synthesises the strengths identified in the literature and is outlined by outlining open research challenges, including concept drift, explainability, adversarial robustness, and privacy-preserving cross-institutional learning.
Abstract
Digital payment platforms now process trillions of transactions annually, and the volume of associated fraud has grown in step with this expansion, with global card-fraud losses estimated in the tens of billions of dollars per year. Rule-based fraud controls, once the industry standard, are increasingly unable to keep pace with adaptive fraudsters and the scale of modern transaction streams. This paper presents a systematic literature review of machine learning (ML) approaches to fraud detection in digital payment systems. Drawing on peer-reviewed studies published between 2016 and 2025, the review organises the field into four broad technique families - classical supervised learning, deep and sequence-based learning, unsupervised and hybrid anomaly detection, and ensemble/reinforcement-learning architectures - and examines the datasets, evaluation metrics, and class-imbalance strategies that recur across the literature. The review finds that while deep sequential models and hybrid ensembles report the strongest detection accuracy on benchmark datasets, most published work still relies on a small number of public, heavily anonymised datasets, limiting generalisability to live payment environments. The paper concludes by proposing a conceptual, layered fraud-detection framework that synthesises the strengths identified in the literature and by outlining open research challenges, including concept drift, explainability, adversarial robustness, and privacy-preserving cross-institutional learning.
Keywords: fraud detection, digital payments, machine learning, deep learning, anomaly detection, credit card fraud, fintech security
In modern high-volume payment systems, detecting fraud is still essentially confined by abhorrent class imbalance, changing transaction patterns, and lack of dependably labelled fraud occurrences. The current research questions the issue of whether self-supervised learning (SSL) can add to the extraction of the knowledge related to fraud in comparison with the capability of strong supervised baselines in the multi-channel payment setting. Using a real world banking dataset of over 13.3 million transactions in the 2010-2019 period, we perform an extensive analysis, including supervised machine learning, anomaly-based SSL, and methods of integrating knowledge into machine learning strategies. Gradient-boosting models are able to build a strong base (F1 = 0.86, ROC-auc = 0.99) that suggests that the trained model has a near-saturation discriminative ability that is solely based on tabular transaction characteristics. We show that naive, generic, SSL-based anomaly detectors lead to reduced precision, and task-adapted representations of supervised models, stacked with task-adapted representations, can increase fraud recall by up to 4.9 with a small F1 increase ( +0.6). However, with strict operationally imposed conditions of accuracy ≥ 0.90, the added benefits of the use of SSL are not experienced, highlighting inherent thresholds of representation based improvement. Such results enhance a knowledge based perspective of when self-supervised representations add value to the decision making and when supervised models have already acquired adequate information about fraud meaning thereby guiding the design of financial fraud knowledge-management models in a robust way.
Boumedyen Shannaq, N. Elshaiekh, Basel Bani-Ismail et al.· ITEGAM- Journal of Engineeri...· 0 citations
This study reviews the development and application of artificial intelligence (AI) and big data analytics (BDA) for fraud detection in accounting and auditing. The review adopts a systematic literature review approach guided by PRISMA principles and synthesizes 20 peer-reviewed and scholarly sources covering data mining, machine learning, natural language processing, deep learning, audit analytics, and big data. The literature indicates that AI and BDA extend fraud detection from periodic, sample-based procedures toward continuous, risk-oriented analysis of large volumes of structured and unstructured data. Machine learning methods, including logistic regression, support vector machines, decision trees, ensemble methods, neural networks, and deep learning, are increasingly used to classify suspicious observations and identify nonlinear fraud patterns. BDA strengthens these models by integrating financial ratios, transaction records, audit evidence, textual disclosures, management commentary, and external information. The review also identifies persistent challenges involving class imbalance, data quality, explainability, privacy, model bias, cybersecurity, and auditor competencies. Overall, the evidence suggests that AI and BDA are most effective when deployed as decision-support mechanisms that complement professional skepticism and audit judgment rather than replace them. Future research should emphasize multimodal data integration, explainable AI, real-time analytics, robust validation across jurisdictions, and governance frameworks for responsible AI-enabled accounting and auditing.
Rosiana Ramadhon, Emmarani Nuristya, Batista Sufa Kefi et al.· Journal of Creative Power a...· 0 citations
Financial institutions process high transaction volumes through digital payment channels, but conventional rule-based fraud detection systems often fail to adapt to changing fraud patterns and may produce excessive false alerts. This study developed and evaluated an intelligent real-time stacked ensemble prototype for credit card fraud detection in financial institutions. The system combined supervised fraud probability estimates from Random Forest and XGBoost with anomaly score evidence from Isolation Forest, and used Logistic Regression as a meta-classifier to generate the final transaction-level decision. The prototype was implemented in Python using Pandas, NumPy, scikit-learn, imbalanced-learn, XGBoost, Seaborn, Matplotlib, and Joblib. The Credit Card Fraud Detection dataset was used for experimentation, containing 284,807 transactions, 30 input features, and 492 fraudulent cases. The data were split using stratified 80:20 sampling; Time and Amount were standardised using training set statistics only, and SMOTE was applied only to the training partition to prevent test-set leakage. The held-out test set retained the original imbalanced distribution of 56,864 legitimate transactions and 98 fraudulent transactions. Random Forest achieved the highest precision and F1-score, with precision of 0.8454, recall of 0.8367, F1-score of 0.8410, false positive rate of 0.00026, and AUC-ROC of 0.9731. XGBoost achieved the highest base-model recall of 0.8673 and AUC-ROC of 0.9750, but produced 136 false positives. Isolation Forest achieved AUC-ROC of 0.8532 but failed to detect fraudulent transactions as a standalone binary classifier under the implemented configuration. The stacked ensemble achieved accuracy of 0.9993, precision of 0.7778, recall of 0.8571, F1-score of 0.8155, false positive rate of 0.00042, and AUC-ROC of 0.9758. The results show that the stacked ensemble produced a usable prototype-level fraud decision layer by combining supervised and anomaly based evidence, although threshold calibration, explainability, and validation on local institutional data remain necessary before operational deployment.
Nwadike U. S., Emmah V. T., M. D.· Journal of Artificial Intell...· 0 citations
Purpose: Public sector financial systems process large volumes of payments, procurement transactions and grants and remain exposed to fraud, corruption and error. Traditional audit-based controls detect only a small share of irregular transactions and usually do so late. This study examines whether machine learning can improve detection of fraudulent transactions in public sector financial systems and compares model families under conditions resembling real government data, including severe class imbalance and scarce fraud labels. Design/methodology/approach: The study builds a synthetic public-sector transaction dataset of 60,000 records that reproduces documented properties of government payment data, including low fraud prevalence, vendor concentration and period-end clustering. Four models are trained and compared: logistic regression, random forest, XGBoost and isolation forest. Models are evaluated using precision, recall, F1 score, area under the receiver operating characteristic curve and area under the precision-recall curve, since accuracy alone misleads under imbalance. Findings: XGBoost achieved the strongest balance of precision and recall (F1 = 0.762, AUC-ROC = 0.992), followed by random forest (F1 = 0.748). Logistic regression achieved high recall but low precision, producing many false alerts. Isolation forest, which is an unsupervised method, performed reasonably well without labelled fraud, which matters since most agencies lack reliable labels. Amount, vendor age, single-bidder procurement and prior irregularity counts were the strongest predictors. Originality/value: This study contributes a reproducible, coded evaluation framework for public sector fraud analytics and gives practitioners evidence-based guidance on model choice under label scarcity and class imbalance, which are conditions that dominate real government settings.
Keywords: Fraud Detection, Machine Learning, Public Sector, Financial Systems, Public Procurement, Anomaly Detection, Government Auditing, XGBoost.
Emmanuel Afriyie Okyem, Robert Kwasi Aboagye Mensah Yeboah, T. Adukpo· Finance & Accounting Res...· 0 citations
This systematic review analyzes 20 influential peer-reviewed studies from 2024 to 2025 and outlines priority directions to move beyond a narrow focus on accuracy toward fraud detection solutions that are scalable, secure, transparent, and cost-effective, enabling more confident deployment of AI-based systems in real-world banking.
Hamid Banirostam, E. Shamsinejad· FinTech and Sustainable Inno...· 0 citations
Financial transaction fraud is a significant problem in the fields of digital banking, credit card transactions, online payment, and mobile financial services. This study systematically reviewed the existing literature on supervised machine learning techniques used for financial fraud detection. The review included peer-reviewed papers published between 2020 and 2026 and followed the PRISMA framework, with Parsifal used to assist the search management, screening, eligibility assessment, and data extraction. A total of 503 records were obtained from IEEE Xplore, Scopus, Web of Science, ScienceDirect, and ACM Digital Library. After removing 129 duplicates, screening 374 records, and assessing 107 full-text articles, a total of 55 studies were accepted in the final review. The findings indicated that financial fraud detection relied more on supervised learning which was the method of choice, especially in cases where labelled data of fraudulent and legitimate transactions were known. Credit/debit card fraud was the most common form of fraud, contributing to 40 studies (72.7%) of the reviewed literature. Methodologically, ensemble, boosting, and hybrid classifiers were the most common methods, occurring in 17 studies (30.9%), followed by deep/hybrid representation learning in 12 studies (21.8%) and classical supervised machine learning comparisons in 10 studies (18.2%). The review further indicated that the model performance not only depends on the algorithm choice but also on preprocessing, feature selection, class imbalance handling, and evaluation metrics. The most commonly reported metric was accuracy, which occurred in 52 studies (94.5%), whereas recall/sensitivity was present in 50 studies (90.9%), precision was in 48 studies (87.3%), F1-score was in 47 studies (85.5%), ROC-AUC/AUC was in 42 studies (76.4%), and confusion matrix was in 38 studies (69.1%). The study ends by recommending that effective supervised fraud detection is a complete methodological pipeline and not just isolated algorithm comparison.
Blessing Bologi, Ridwan Kolapo, Temitope Olufunmi Atoyebi· International journal of res...· 0 citations
We use cookies to run the site and, with your consent, for analytics and to show ads.
See our Cookie Policy.
