Leveraging Predictive Analytics to Strengthen Financial Oversight in Government Expenditure: A Case for Public Sector Reform

 Corresponding author: Herbert Otim Copyright © 2025 Author(s) retain the copyright of this article. This article is published under the terms of the Creative Commons Attribution Liscense 4.0. Leveraging Predictive Analytics to Strengthen Financial Oversight in Government Expenditure: A Case for Public Sector Reform Herbert Otim 1, *, Mercy Elizabeth Arinda 2 and Frank Appiah-Oware 2 1 Department of Information Technology and Analytics, Kogod School of Business, American University, Washington D.C., United States of America. 2 Department of Accounting, Kogod School of Business, American University, Washington D.C., United States of America. World Journal of Advanced Research and Reviews, 2025, 28(01), 298-314 Publication history: Received on 21 August 2025; revised on 01 October 2025; accepted on 03 October 2025 Article DOI: https://doi.org/10.30574/wjarr.2025.28.1.3389 Abstract Governments increasingly seek to strengthen transparency and accountability in public financial management, yet traditional, retrospective audits struggle to surface irregularities at the speed and scale of modern procurement. This study develops and applies a practical analytics framework to U.S. Department of Commerce (DOC) procurement transactions for FY2025, demonstrating how unsupervised learning can triage large award corpora into tractable, auditsalient subsets. Using 16,581 transactions from the USAspending Award Data Archive, we engineer features aligned to established risk theories: approval lag (solicitation-to-action timing), vendor history (prior awards and concentration), award magnitude (obligations, base-and-options, potential ceilings), and award structure/competition (IDV relationships, pricing type, and extent competed)—and apply Isolation Forest (contamination = 1%). The model flags 166 atypical transactions (≈1%) characterized by (i) extreme potential award ceilings (median ≈ $8B), (ii) orderdependent pricing under Indefinite Delivery Vehicles (IDVs), and (iii) competition pathways reported as “full and open after exclusion of sources.” Sensitivity analysis shows anomalies are highly threshold-dependent (0–2% contamination yields 0–≈330 flags), underscoring the need to calibrate cutoffs to investigative capacity. While findings are not determinations of non-compliance, they delineate priority cases for follow-up testing (e.g., ceiling-to-obligation reconciliation, order-level pricing documentation, justification memos for exclusions). The framework translates directly to oversight practice via score-band triage, dashboarding, and model governance (documentation, fairness checks, periodic recalibration). Limitations include the absence of ground-truth labels and potential measurement error in administrative data; future work should integrate supervised models (e.g., logistic/ensemble learners) using adjudicated outcomes and employ explainability techniques to attribute anomaly drivers. Overall, results illustrate that predictive analytics can complement audits, reduce detection lag, and inform evidence-based policy within public procurement systems. Keywords: Predictive Analytics; Anomaly Detection; Procurement Oversight; Government Expenditure; Financial Accountability; Public Sector Reform 1. Introduction Public financial management (PFM) is a central pillar of effective governance. It determines how public resources are mobilized, allocated, and accounted for in pursuit of social and economic objectives. Yet across both developed and developing economies, governments continue to face persistent challenges in achieving transparency, accountability, and efficiency in expenditure management. Procurement, which accounts for a substantial portion of government spending, has been identified by the Organisation for Economic Co-operation and Development (OECD) as one of the most corruption-prone and audit-sensitive functions of the state [4,13,25]. The U.S. Government Accountability Office World Journal of Advanced Research and Reviews, 2025, 28(01), 298-314 299 (GAO) estimates that federal agencies made $247 billion in improper payments in FY2022 alone [7], underscoring the urgent need for more effective oversight mechanisms. 1.1. Limitations of traditional audits Traditional auditing methods have long served as the backbone of accountability systems. These approaches rely on retrospective review and sample-based testing, and they continue to provide valuable assurance. However, they also exhibit structural limitations. First, they are inherently reactive: by the time anomalies are detected, resources may already have been disbursed or misallocated. Second, sample-based approaches may fail to capture systemic irregularities hidden in large data populations. Finally, manual audits are resource-intensive and slow, constraining the ability of oversight bodies to detect risks in real time [1,2,6]. 1.2. Predictive analytics as a complement Recent advances in predictive analytics and machine learning offer a modern complement to these limitations. Predictive models can examine entire populations of transactions rather than samples, allowing governments to identify unusual patterns, classify transactions by risk, and optimize resource allocation for investigations. Methods such as anomaly detection, process mining, and classification models have been widely applied in financial services and privatesector auditing, where they have proven effective in uncovering fraud, waste, and inefficiencies [9,10,16,18]. Embedding similar approaches into government procurement oversight could transform accountability by shifting from detectionafter-the-fact to prevention and early warning [3,7,14]. 1.3. Policy context and international guidance The policy environment increasingly favors adoption of analytics. The OECD [4,25] emphasizes the role of digitalization in improving procurement transparency. The IMF [8] argues that digital public financial management is a key lever for improving fiscal discipline. The World Bank [12] advocates for the use of data analytics to strengthen procurement integrity in lowand middle-income countries. In the U.S., the GAO [7,24] has repeatedly highlighted analytics as critical to reducing improper payments and advancing evidence-based governance. Meanwhile, professional services firms such as PwC [23] report that internal audit departments are rapidly expanding their use of data analytics, even though adoption remains uneven across public institutions. 1.4. Research gap Despite strong policy momentum, the application of advanced anomaly detection methods to federal procurement data remains limited. Prior research has explored process mining [1], forensic accounting [2], and data-driven accountability [3], but few studies have systematically tested unsupervised anomaly detection models on large-scale U.S. award data. Moreover, much of the existing literature focuses on retrospective fraud cases rather than proactive risk identification. This creates an opportunity to demonstrate how predictive analytics can be integrated into public sector oversight in a practical, reproducible way. 1.5. Objectives and hypothesis This study addresses that gap by applying machine learning to the FY2025 Department of Commerce (DOC) procurement dataset. Specifically, it tests whether anomaly detection can surface a small, tractable subset of awards for audit triage. Guided by theory and audit practice, the study advances the following hypothesis: H1: Transactions with extended approval lags, high contract values, or inconsistent vendor histories are more likely to be classified as high-risk or non-compliant by anomaly detection models. 1.6. Contributions The paper makes three contributions. First, it provides empirical evidence that anomaly detection can prioritize oversight in U.S. federal procurement, surfacing red flags around Indefinite Delivery Vehicles (IDVs), extreme contract ceilings, and restricted competition pathways. Second, it outlines a reproducible analytics pipeline, including preprocessing, feature engineering, and model calibration, that oversight bodies can adapt. Third, it translates results into actionable policy implications, emphasizing triage rules, governance guardrails, and long-term integration of analytics into audit practice. World Journal of Advanced Research and Reviews, 2025, 28(01), 298-314 300 1.7. Roadmap of the paper The remainder of this paper is organized as follows. Section 2 reviews prior literature on analytics in auditing, procurement oversight, and fraud detection. Section 3 describes the data, preprocessing steps, feature engineering, and anomaly detection methodology. Section 4 presents the results, including descriptive statistics, anomalies flagged, and categorical breakdowns. Section 5 discusses findings in light of existing literature and oversight practice. Section 6 provides a policy roadmap for implementing analytics in government oversight. Section 7 outlines limitations and future research opportunities. Section 8 concludes. 2. Literature Review The literature on predictive analytics in auditing, procurement oversight, and fraud detection has grown rapidly over the past two decades. This section synthesizes findings from four interrelated streams: (i) analytics in auditing and continuous assurance, (ii) procurement oversight and corruption risks, (iii) applications of machine learning in fraud detection, and (iv) the policy and practice context for adopting analytics in public financial management. Together, these streams provide the foundation for applying anomaly detection to U.S. government procurement data. 2.1. Analytics in Auditing and Continuous Assurance Auditing is traditionally retrospective and rule-based, but advances in analytics have increasingly transformed how auditors approach large datasets. Jans, Alles, and Vasarhelyi [1] conducted a field study applying process mining to event logs, demonstrating how inconsistencies in transaction flows can be revealed through data-driven analysis. This early work highlighted the potential for analytics to uncover irregularities that conventional sampling might miss. Alles [5] examined the drivers and obstacles to the adoption of big data in the auditing profession, noting both enthusiasm for risk-focused analytics and hesitation due to resource constraints. Similarly, Vasarhelyi, Kuenkaikaew, and Littley [6] discussed continuous assurance models that leverage technology to enable ongoing compliance monitoring. These models reduce detection lag and allow auditors to respond more quickly to irregularities. Appelbaum, Kogan, and Vasarhelyi [16] identified research needs at the intersection of big data and auditing, emphasizing the importance of integrating structured and unstructured data sources. Their work underlined the necessity of moving beyond traditional financial data to capture operational and contextual signals relevant for assurance. Collectively, these studies underscore that analytics can extend audit coverage, reduce human error, and improve detection of atypical behaviors. 2.2. Government Procurement Oversight and Corruption Risks Public procurement accounts for 12–20% of GDP in many economies, making it one of the most significant areas of government spending [13]. It is also highly vulnerable to corruption, collusion, and inefficiency. The European Court of Auditors [11] has repeatedly flagged weaknesses in EU procurement oversight, emphasizing the potential of big data to support more effective monitoring. The OECD [13] highlights that procurement fraud often arises through restricted competition, inflated contract ceilings, and opaque contractual arrangements. The World Bank [12] stresses the importance of analytics for strengthening procurement integrity, particularly in lowand middle-income countries where governance frameworks may be weaker. By analyzing large volumes of contract data, oversight institutions can detect bid rigging, monitor vendor concentration, and assess whether competition procedures are followed. The GAO has echoed these concerns in the U.S. context. In reports on improper payments and data innovation, GAO [7,14,24] calls for more systematic use of analytics in government oversight. Similarly, the European Commission’s AntiFraud Strategy [15] emphasizes data analysis as a complement to traditional monitoring tools. These policy-oriented contributions consistently highlight procurement as an audit-sensitive domain where analytics can provide outsized benefits. 2.3. Machine Learning in Fraud and Risk Detection Beyond the audit and procurement literature, there is a vast body of research on machine learning for fraud detection in finance, insurance, and e-commerce. Ngai et al. [9] reviewed applications of data mining techniques in financial fraud detection, proposing a classification framework that remains widely cited. They concluded that both supervised and unsupervised approaches are effective, depending on whether labeled outcome data are available. World Journal of Advanced Research and Reviews, 2025, 28(01), 298-314 301 Baesens, Van Vlasselaer, and Verbeke [10] extended these insights by focusing on descriptive, predictive, and social network analytics for fraud prevention. Their work illustrates how anomaly detection and network models can expose collusion and coordinated fraud patterns. Cao and Yu [17] highlighted the role of data mining in business analytics, providing methodological guidance for applications across industries. Pedregosa et al. [18] introduced scikit-learn, the Python machine learning library used in this study. Its implementation of Isolation Forest makes anomaly detection accessible and scalable for high-dimensional datasets. Empirical studies show that Isolation Forest performs well in contexts where fraud or non-compliance outcomes are rare and labels are missing. This makes it a strong fit for government procurement oversight. 2.4. Policy Context: Digital Government and Analytics Finally, a growing policy literature addresses the governance of analytics in public financial management. The OECD [4,19,25] emphasizes digital transformation as a prerequisite for transparent and accountable government, noting that data-driven decision-making improves both efficiency and trust. The IMF [8] similarly stresses digitalization of PFM systems as a means to strengthen fiscal discipline and reduce vulnerabilities. The World Bank [12] has supported pilot programs that integrate analytics into procurement monitoring, reporting positive results in reducing collusion and increasing compliance. PwC [23] reports that internal audit departments across the private and public sectors are adopting analytics at a rapid pace, though challenges remain in terms of skills, data quality, and governance. The GAO [24] frames analytics not only as a technical upgrade but as an accountability imperative, recommending that federal agencies embed data-driven methods into their oversight systems. The European Anti-Fraud Office (OLAF) [22] also underscores the role of data analysis in detecting irregularities in EU spending programs. These reports stress that adoption must be accompanied by governance safeguards, including transparency, fairness checks, and documentation. 2.5. Synthesis and Research Gap Across these four streams, several themes emerge. First, analytics extends the reach of audits by enabling populationlevel testing and early detection of anomalies. Second, procurement is universally recognized as a high-risk domain that stands to benefit from analytics-driven oversight. Third, machine learning methods such as anomaly detection are particularly suited to contexts where fraud is rare and labels are absent. Finally, international policy guidance underscores the urgency of embedding analytics into PFM systems, while cautioning that adoption requires robust governance. Despite this momentum, few empirical studies have systematically applied anomaly detection to U.S. federal procurement data. Existing literature often focuses on retrospective fraud cases or general frameworks, leaving a gap in practical, reproducible applications. This study fills that gap by applying Isolation Forest to DOC FY2025 awards, demonstrating both empirical patterns and practical implications for oversight. 3. Methodology 3.1. Data Source The dataset used in this study was drawn from the U.S. Award Data Archive on USAspending.gov [26], which aggregates federal procurement award information. We focus on the Department of Commerce (DOC) transactions for FY2025, comprising 16,581 unique procurement transactions after initial filtering. Each record contains award identifiers, financial attributes, dates, vendor details, agency information, and categorical descriptors such as contract type, pricing arrangement, and competition pathway. The DOC was chosen as a case study for three reasons. First, it represents a large civilian agency with diverse procurement activity. Second, its portfolio features a high proportion of Indefinite Delivery Vehicles (IDVs), where pricing is determined at the order level, introducing potential risk. Third, Commerce spending patterns reflect both large, high-value awards and routine, small contracts, allowing analytics to differentiate across scales. World Journal of Advanced Research and Reviews, 2025, 28(01), 298-314 302 3.2. Preprocessing 3.2.1. Data cleaning Normalization of headers. Column names were standardized to lowercase using underscores to ensure consistency across the dataset. This step simplified subsequent data handling and minimized the risk of referencing errors during preprocessing. Type coercion. Several fields, including contracting officer names, contained mixed data types that triggered warnings on load. These were coerced to strings to maintain uniformity and prevent disruptions in feature encoding. Handling missing values. Entirely missing fields, such as Department of Defense–specific acquisition program codes, were excluded from the analysis to reduce noise and avoid distortions in the feature matrix. For variables with partial missingness, imputation methods were applied to preserve data integrity. Specifically, numeric fields were imputed using median values, while categorical fields were imputed using modal categories. This approach ensured that the dataset remained complete and analyzable without introducing strong distributional biases. Consistency checks. Contract dates were validated to ensure logical sequencing between solicitation and action. Specifically, each record was examined to confirm that the solicitation date preceded or was equal to the action date. This step safeguarded against data entry errors and ensured the temporal integrity of the dataset. 3.2.2. Multicollinearity analysis Variance Inflation Factor (VIF) scores were calculated for core financial predictors (e.g., obligations, base-and-options, potential ceilings). All values fell between 1.0 and 1.13, indicating low linear redundancy and confirming that the feature matrix was well-conditioned for anomaly detection. 3.3. Feature Engineering Feature construction. Feature engineering was guided by the study’s central hypothesis that extended approval lags, high contract values, and vendor inconsistency increase compliance risk. An approval lag variable was created by measuring the number of days between solicitation_date and action_date. This feature captures potential delays in award processing, which may indicate oversight or procedural inefficiencies. To assess vendor history, three metrics were derived: the count of prior awards, cumulative obligations, and a vendor concentration index representing each vendor’s share of total Commerce obligations. These measures highlight dependence on particular suppliers and help detect unusual vendor dominance. Financial magnitude variables were included to represent obligations, base-and-options values, and potential award ceilings. Outliers were preserved deliberately, as anomaly detection techniques rely on identifying extreme observations. Finally, contract structure indicators were encoded to reflect IDV usage, pricing type, and competition pathways, while categorical data such as NAICS industry codes and vendor states were transformed using one-hot encoding. After these transformations, the feature matrix contained 594 engineered variables. 3.4. Model and Rationale The Isolation Forest (iForest) algorithm was selected for anomaly detection. Originally introduced by Liu et al., it isolates anomalies by constructing random decision trees. Points requiring shorter path lengths are more easily isolated and therefore more likely to be anomalous. The anomaly score is computed as: 𝑠(𝑥,𝑛)= 2−𝐸(ℎ(𝑥)) 𝑐(𝑛) where E(h(x)) is the average path length for instance x across the forest, and c(n) is the expected path length for a dataset of size n. Rationale for model choice. The Isolation Forest algorithm was selected for anomaly detection because it combines efficiency and interpretability in high-dimensional contexts. First, it is highly scalable, making it suitable for the DOC World Journal of Advanced Research and Reviews, 2025, 28(01), 298-314 303 dataset of 16,581 records and 594 engineered features. Second, it is unsupervised, which is essential given the absence of labeled fraud outcomes in government procurement data. Third, its outputs are easily interpretable: anomaly scores fall between 0 and 1, allowing transactions to be rank-ordered for audit triage. Alternative models such as one-class SVM and DBSCAN were evaluated but found to be less scalable or overly sensitive to parameter tuning in highdimensional data, making Isolation Forest the most practical choice for this study. 3.5. Parameterization The model was implemented with 200 estimators (trees) and a subsample size of 256, consistent with the scikit-learn defaults. The contamination parameter was set to 0.01, meaning 1% of transactions were expected to be anomalous. This setting aligned with the study’s expectation that only a small subset of awards would represent compliance risks. A fixed random state of 42 was used to ensure reproducibility of results. Notably, the contamination parameter required explicit tuning: using the “auto” option produced no anomalies in this dataset, underscoring the importance of parameter selection for effective application. 3.6. Sensitivity Analysis To assess robustness, sensitivity analysis was conducted across different contamination thresholds. When the contamination parameter was set to “auto,” the model flagged no anomalies, indicating that the default threshold was too conservative for this dataset. At 0.5% contamination, approximately 80 anomalies were identified. Increasing the parameter to 1% produced 166 anomalies, while a further increase to 2% surfaced roughly 330 anomalies. The 1% setting provided the most plausible balance between surfacing meaningful anomalies and minimizing noise, consistent with expectations that only a small fraction of transactions are likely to represent compliance risks. 3.7. Validation and robustness Because no ground-truth fraud labels exist, validation was qualitative. The flagged anomalies were profiled against descriptive statistics and categorical breakdowns. The interpretability of clusters (e.g., extreme ceilings, IDV pricing, exclusionary competition) served as a form of face validity. Future research (see Section 7) proposes integrating labeled outcomes (e.g., prior questioned costs, audit exceptions) for supervised model validation. 4. Results and Discussion 4.1. Anomaly Detection Outcomes Figure 1 Distribution of anomaly scores for DOC FY2025 awards. At a contamination setting of 1%, the Isolation Forest assigned anomaly scores that were tightly clustered around a mean of 0.166 (SD ≈ 0.011). Scores were bounded between 0 and 1, with higher values indicating a greater likelihood of anomalous behavior. The resulting distribution (Figure 1) was skewed, with most inlier records exhibiting scores World Journal of Advanced Research and Reviews, 2025, 28(01), 298-314 304 below 0.17, while the 166 flagged anomalies concentrated within the narrower range of 0.18–0.22. This pattern indicates that the anomalies were not random noise but rather a statistically distinct subset of transactions warranting closer attention. 4.2. Financial Characteristics Anomalous transactions were distinguished primarily by their extreme award magnitudes. The median potential award ceiling among anomalies was approximately $8 billion, compared to only $0.24 million for normal transactions. In addition, several anomalies displayed negative values for the “all-options” measure, indicating either de-obligations or unusual contract modifications. While such adjustments are not inherently improper, they represent atypical financial patterns that warrant closer scrutiny. Table 1 Summary statistics comparing anomalies and normal transactions (in millions). Normal Descriptive Statistics Metric Anomalies Normals Count 166 16,415 Mean potential value $7.03B $0.5M Median potential value $8B $240K Std. dev. (potential) High (~±1.2B) Moderate Mean anomaly score 0.166 0.128 Figure 2 Distribution of potential award values (log-scaled). Figure 2 shows the stark divergence between anomalies and normal transactions: while most DOC awards cluster under $1M, anomalies spike at multi-billion-dollar ceilings, creating a bimodal distribution. 4.3. Categorical Characteristics Categorical features reveal clear clustering of anomalies in specific procurement categories. World Journal of Advanced Research and Reviews, 2025, 28(01), 298-314 305 4.3.1. Contract pricing types A striking pattern emerged when examining contract pricing structures. Approximately 90% of anomalous transactions were classified under order-dependent pricing, where Indefinite Delivery Vehicles (IDVs) allow pricing arrangements to be determined separately for each order. By contrast, the majority of normal awards followed more conventional structures such as fixed-price or cost-reimbursement contracts. This overrepresentation of order-dependent pricing among anomalies suggests that such arrangements may create vulnerabilities by deferring key cost decisions and reducing immediate transparency. Figure 3 Distribution of contract pricing types among anomalies. This overrepresentation suggests a structural vulnerability: IDVs with deferred pricing decisions may allow inflated ceilings without immediate cost justification. 4.3.2. Extent competed Of the 166 anomalies, 90 were awarded under “Full and open competition after exclusion of sources.” A smaller subset (≈12) fell under simplified acquisition procedures (SAP) or were not competed. World Journal of Advanced Research and Reviews, 2025, 28(01), 298-314 306 Figure 4 Distribution of extent competed for anomalies The clustering in exclusionary competition pathways is consistent with OECD [13] warnings that restricted competition is a hallmark of procurement risk. 4.3.3. Vendor geography Anomalous awards were disproportionately associated with vendors registered in Virginia (VA) and Maryland (MD). While these states are logical hubs due to their proximity to federal contracting activities, the concentration of anomalies within such a narrow geographic base raises potential concerns. This clustering may reflect structural dependencies on a limited set of vendors, which could reduce competition and increase the risk of collusion or repeated irregularities within the procurement ecosystem. Figure 5 Top vendor states for anomalies. Vendor concentration may increase risks of collusion or reduced competition, particularly if award structures consistently favor the same regions. World Journal of Advanced Research and Reviews, 2025, 28(01), 298-314 313 Statement of ethical approval This study did not require ethical approval as it utilized publicly available secondary data from the USAspending.gov archive, with no involvement of human or animal subjects. Statement of informed consent Not applicable. This research did not involve human participants. References [1] Jans M, Alles M, Vasarhelyi MA. A field study on the use of process mining of event logs as an analytical procedure in auditing. The Accounting Review. 2014;89(5):1751–73. doi:10.2308/accr-50807. [2] Kranacher MJ, Riley RA, Wells JT. Forensic Accounting and Fraud Examination. Hoboken: Wiley; 2019. [3] Rao HR, Sridhar S. Exploring the role of analytics in government accountability. Journal of Public Administration Research and Theory. 2018;28(1):1–14. doi:10.1093/jopart/mux003. [4] OECD. Public Financial Management and Digitalisation. Paris: OECD Publishing; 2022. doi:10.1787/9789264303419-en. [5] Alles MG. Drivers of the use and facilitators and obstacles of the evolution of Big Data by the audit profession. Accounting Horizons. 2015;29(2):439–49. doi:10.2308/acch-51067. [6] Vasarhelyi MA, Kuenkaikaew S, Littley J. Continuous assurance and the use of technology for business compliance. Newark: Rutgers Business School; 2012. [7] U.S. Government Accountability Office (GAO). A Framework for Managing Improper Payments in Federal Programs. Washington, DC: GAO; 2016. Available from: https://www.gao.gov/products/gao-16-236. [8] International Monetary Fund (IMF). Digitalization for Public Financial Management. Washington, DC: IMF; 2021. Available from: https://www.imf.org/en/Publications/Staff-DiscussionNotes/Issues/2021/06/24/Digitalization-for-Public-Financial-Management. [9] Ngai EW, Hu Y, Wong YH, Chen Y, Sun X. The application of data mining techniques in financial fraud detection: A classification framework and an academic review. Decision Support Systems. 2011;50(3):559–69. doi:10.1016/j.dss.2010.08.006. [10] Baesens B, Van Vlasselaer V, Verbeke W. Fraud Analytics: Using Descriptive, Predictive, and Social Network Techniques to Detect and Prevent Fraud. Hoboken: Wiley; 2015. [11] European Court of Auditors. Use of Big Data in EU Audit. Luxembourg: Publications Office of the EU; 2019. Available from: https://op.europa.eu/en/publication-detail/-/publication/58bcded9. [12] World Bank. Enhancing Government Transparency Through Data Analytics. Washington, DC: World Bank; 2020. Available from: https://documents.worldbank.org. [13] OECD. Preventing Corruption in Public Procurement. Paris: OECD Publishing; 2017. doi:10.1787/9789264274198-en. [14] U.S. Government Accountability Office (GAO). Data Analytics Innovation in Oversight. Washington, DC: GAO; 2021. Available from: https://www.gao.gov/products/gao-21-498. [15] European Commission. Anti-Fraud Strategy for EU Spending. Brussels: European Commission; 2020. Available from: https://ec.europa.eu/anti-fraud. [16] Appelbaum D, Kogan A, Vasarhelyi MA. Big data and analytics in the modern audit engagement: Research needs. Auditing: A Journal of Practice & Theory. 2017;36(4):1–27. doi:10.2308/ajpt-51707. [17] Cao L, Yu PS. Data Mining for Business Analytics: Concepts, Techniques, and Applications in Python. Cham: Springer; 2016. doi:10.1007/978-3-319-44672-0. [18] Pedregosa F, Varoquaux G, Gramfort A, Michel V, Thirion B, Grisel O, et al. Scikit-learn: Machine learning in Python. Journal of Machine Learning Research. 2011;12:2825–30. Available from: https://jmlr.org/papers/v12/pedregosa11a.html. [19] OECD. Government at a Glance 2019. Paris: OECD Publishing; 2019. doi:10.1787/gov_glance-2019-en. World Journal of Advanced Research and Reviews, 2025, 28(01), 298-314 314 [20] Kauffman RJ, Tsai JY. The impact of fraud detection systems on online consumer behavior. Electronic Commerce Research and Applications. 2010;9(3):191–202. doi:10.1016/j.elerap.2009.05.001. [21] Appelbaum D, Nehmer RA. Using drones in audit: Opportunities and challenges. International Journal of Accounting Information Systems. 2017;27:1–14. doi:10.1016/j.accinf.2017.08.001. [22] European Anti-Fraud Office (OLAF). Annual Report 2020. Brussels: European Commission; 2020. Available from: https://anti-fraud.ec.europa.eu. [23] PricewaterhouseCoopers (PwC). State of Internal Audit Analytics. London: PwC; 2020. Available from: https://www.pwc.com. [24] U.S. Government Accountability Office (GAO). Opportunities to Strengthen Federal Program Oversight With AI and Analytics. Washington, DC: GAO; 2022. Available from: https://www.gao.gov/products/gao-22-105876. [25] OECD. Digital Government Review of Sweden: Towards a Data-Driven Public Sector. Paris: OECD Publishing; 2020. doi:10.1787/4de9f5bb-en. [26] USAspending.gov. Award Data Archive. Washington, DC: U.S. Department of the Treasury; 2025. Available from: https://www.usaspending.gov/download_center/award_data_archive.