A Privacy-Preserving Federated Learning Framework For Attrition And Engagement Prediction In SAP Success Factors

20 Mar
Posted on Author

Authors: Liang Zhao, Seojun Lee, Haruto Saito, Robert Anderson

Abstract: The increasing reliance on data-driven decision making in human capital management has intensified the need for predictive analytics models that can accurately forecast employee attrition and engagement while preserving data privacy across organizational boundaries. Traditional machine learning approaches in SAP SuccessFactors environments typically require centralized data aggregation, which introduces significant risks related to data exposure, regulatory non-compliance, and cross-border data transfer limitations. This study proposes a privacy-preserving federated learning architecture designed to enable collaborative workforce analytics across distributed SAP SuccessFactors tenants without the need to share raw employee data. The framework leverages decentralized model training, where local models are trained within individual tenant environments and only encrypted model updates are shared with a central aggregation layer. Advanced privacy-enhancing techniques, including secure aggregation, differential privacy, and role-based access controls, are incorporated to ensure confidentiality and compliance with global data protection standards. The proposed architecture is further enhanced through integration with SAP Business Technology Platform services, enabling scalable orchestration, real-time model updates, and seamless interoperability with existing HR modules such as Employee Central and Performance Management. Experimental evaluation using simulated multi-tenant datasets demonstrates that the federated approach achieves predictive performance comparable to centralized models while significantly reducing privacy risks and data governance challenges. The results highlight improvements in model generalization, reduced bias across organizational units, and enhanced trust in AI-driven decision support systems. This research contributes a novel, enterprise-ready framework for secure, scalable, and privacy-aware workforce analytics, positioning federated learning as a viable solution for next-generation intelligent HR systems within SAP SuccessFactors landscapes.

DOI: https://doi.org/10.5281/zenodo.19128614