A Data-Driven Framework for Detecting Procurement Anomalies and Financial Control Risks in Service Organisations Using Machine Learning and Risk-Based Analytics

Authors

  • Mary Jatta

    School of Business, San Francisco Bay University, California, USA
    Author
  • Sarah Cobbinah

    School of Business, San Francisco Bay University, California, USA
    Author
  • Alimatou Jallow

    American International University West Africa, Serekunda, Gambia
    Author
  • Veronica Ibe

    Department of Microbiology, Lagos State University, Lagos, Nigeria
    Author
  • Toluwalase Ayeni

    Department of Economics, University of Ilorin, Ilorin, Nigeria
    Author
  • Quadri Salami

    Department of Electrical and Electronics Engineering, University of Lagos, Nigeria
    Author

DOI:

https://doi.org/10.4314/

Abstract

The fact that purchasing activities are a major source of financial exposure in service organizations arises because procurement transactions cover a large number of vendors, involve various invoices, purchase orders, approvals, and contract-related decisions. Control procedures that rely mainly on periodic sampling are unlikely to pick up complex, scattered, or quickly changing anomalies. In this study, a data-driven approach is presented for detecting purchasing anomalies and financial control risks by combining statistical red-flag analysis, anomaly detection, machine learning, and risk-based audit prioritization. The methodology was created through a secondary analysis of actual procurement cases, comprising about 216,000 procurement transactions from a large government organization in Singapore, over 50,000 procurement observations from SF Express, and around 3.5 million electronic invoices with more than 11 million line items from public bodies in Brazil. The framework in question brings together transaction-value outliers, unusual purchasing frequencies, duplicate or near-duplicate invoices, vendor concentration, abnormal price deviations, temporal patterns, approval exceptions, and Benford-law-based indicators to form a composite purchasing-risk score. Existing evidence shows the practical usefulness of this method: an explainable procurement anomaly-detection system that used 48 fraud indicators reached 67.1% precision in identifying suspicious transactions in the Singapore example, while a machine-learning study based on more than 50,000 procurement records proved the possibility of automatically classifying procurement fraud. Further analysis of the Brazilian electronic invoices revealed that price alerts of high intensity amounted to more than R$560 million in invoice-item value. The results show that data analytics can shift purchasing controls from being mainly based on retrospective sampling to continuous, risk-based monitoring. Nevertheless, the anomaly scores should be seen as indicators that need to be investigated by humans and not as proof of fraud. The study concludes that by combining machine learning, statistical analysis, explainability, and well-established internal-control principles, a practical foundation is provided for improving financial control in service organizations.

Author Biographies

  • Mary Jatta, School of Business, San Francisco Bay University, California, USA

     

     

  • Sarah Cobbinah, School of Business, San Francisco Bay University, California, USA

     

     

  • Alimatou Jallow, American International University West Africa, Serekunda, Gambia

     

     

     

     

  • Veronica Ibe, Department of Microbiology, Lagos State University, Lagos, Nigeria

     

     

     

     

     

  • Toluwalase Ayeni, Department of Economics, University of Ilorin, Ilorin, Nigeria

     

     

     

  • Quadri Salami, Department of Electrical and Electronics Engineering, University of Lagos, Nigeria

     

     

     

Downloads

Published

2023-09-19

How to Cite

A Data-Driven Framework for Detecting Procurement Anomalies and Financial Control Risks in Service Organisations Using Machine Learning and Risk-Based Analytics. (2023). Communication In Physical Sciences, 9(4), 1247-1263. https://doi.org/10.4314/