Privacy-Preserving Machine Learning Techniques for Cross-Sector Fraud Detection in the South African Financial and Telecommunications Sectors
Lufungula Osembe ORCID ↗ · Mukosi Lucky Mashudu ORCID ↗
International Journal of Business Ecosystem & Strategy · 2026 · Vol. 8 / Issue 4 · 244-267
Publication metrics
External links open provider records or DOI searches. Provider counts are not combined. Mendeley readership measures saves, not citations.
Dimensions
Live provider badge. If unavailable, use the Dimensions link above.
Observed 21 Oct 2026, 00:14 UTC
Observed 21 Oct 2026 00:14 UTC
OpenAlex and Mendeley observations refresh on article visits, at most once per 24 hours. Unavailable data is not zero. Google Scholar opens a DOI search.
Citations received by this article, grouped by the year of the citing work. Crossref and OpenAlex lifetime totals remain separate in Publication metrics.
2017–2026 Annual history last updated:
Citations received
OpenAlex · year of the citing work
Hover over a year or focus the chart and use arrow keys to inspect values.
2017–2026 annual-history window. 2026 is year to date. Unknown years have gaps.
View data & coverage
2026 is incomplete. Within a successfully fetched ten-year window, omitted years mean zero. Years beyond a saved observation’s window remain unknown (—). Older citations are excluded, so this chart’s sum may differ from the OpenAlex lifetime total. The date above belongs to this annual history; an unsuccessful refresh keeps the previous observation.
Abstract
Cross-sector fraud between South Africa’s financial and telecommunications sectors has intensified, driven by sophisticated criminal networks that exploit operational disparities between these industries. Traditional fraud detection systems, which depend on centralised data sharing and siloed organisational structures, encounter substantial legal and practical obstacles under the Protection of Personal Information Act (POPIA). These include restrictions on cross-entity data transfer, data minimisation requirements, and transparency obligations for automated decision-making. This study developed and evaluated a privacy-preserving framework that integrates Federated Learning (FL) and Explainable Artificial Intelligence (XAI) to enable collaborative fraud detection while safeguarding customer privacy and ensuring regulatory compliance. A mixed-method approach was employed. Quantitative experiments were conducted on a dataset of 35,675 credit card transactions linked to telecommunications parameters, training a federated global model across simulated financial and telecommunications providers. A qualitative systematic literature review identified key barriers to cross-sector collaboration and distilled lessons from prior Federated Learning and Explainable Artificial Intelligence implementations. The federated model demonstrated strong detection performance, achieving a ROC-AUC of 0.98, near-perfect precision and recall for legitimate transactions, and an F1-score of 0.89 for the minority fraud class, all without exchanging raw customer data between institutions. SHAP analysis identified TransactionAmt and card1 as the most influential features, thereby satisfying POPIA's transparency mandates. While federated approaches did not consistently surpass siloed models in raw accuracy, this marginal trade-off reflects the essential privacy-utility equilibrium required for legal compliance. The framework effectively reconciled collective security with individual privacy, confirming that POPIA adherence need not diminish detection efficacy. This study provides a validated blueprint for financial and telecommunications providers in South Africa to collaboratively counter cross-sector fraud within regulatory constraints. The framework further supports audit trail generation and decision rationales, addressing compliance obligations. Drawing on systematic literature review insights, particularly regarding collaboration barriers and prior Federated Learning and Explainable Artificial Intelligence implementations, the findings offer regulators and investors practical guidance for meeting statutory requirements without compromising privacy rights.
Keywords: {'name': 'Federated Learning'}, {'name': 'Machine Learning'}, {'name': 'Financial and Telecommunication sectors'}, {'name': 'Fraud Detection'}, {'name': 'Expandable AI'}
Authors & affiliations
Lufungula Osembe
The Independent Institute of Education
Mukosi Lucky Mashudu
The Independent Institute of Education
Deposited references
Supplied references, not a list of publications citing this article.
No reference list supplied.