Federated Predictive Analytics in Finance: Privacy-Preserving Churn and Fraud Forecasting Across Institutions
Summary
No summary yet. VIRAT can read this paper and write a plain-language summary of its aim, methods, key findings and limitations.
Summarise with VIRAT Free to try with your Science Catalogs account.Abstract
Abstract The escalating adoption of machine learning in financial services is impeded by regulatory constraints on data sharing and the fragmentation of customer data across competing institutions. This article investigates how federated learning (FL) architectures—specifically FedProx with differential privacy stochastic gradient descent (DP-SGD) and secure aggregation—can enable collaborative churn and fraud prediction without the transfer of raw data. Using simulated multi-institutional banking benchmarks across five institutions (2025–2026), our experiments demonstrate that federated models achieve AUC-ROC improvements of 1.2–2.4 percentage points over single-institution baselines, while maintaining privacy budgets below ε = 4.5 and reducing demographic parity gaps by approximately 48–49% relative to pre-federation levels. Regulatory alignment with GDPR Article 25, CCPA, and DORA's operational resilience mandates is assessed using structured compliance audits. Our findings indicate that post-2025 federated architectures offer a viable pathway for collaborative financial analytics, substantially improving predictive performance and fairness while preserving institutional data sovereignty. Keywords: federated learning, churn prediction, fraud detection, differential privacy, secure aggregation, regulatory compliance, bias reduction, financial analytics, GDPR, DORA
Fields of Study & Subjects
Privacy-Preserving Technologies in Data
Customer churn and segmentation
Blockchain Technology Applications and Security
Publication details
- Journal / source
- INTERNATIONAL JOURNAL OF RESEARCH IN MANAGEMENT
- Publication date
- 26 March 2026
- Language
- EN