International Journal of Scientific Research and Engineering Development

Predictive analytics has become a cornerstone in modern data-driven decision-making across domains such as healthcare, finance, smart cities, and e-commerce. However, the increasing reliance on sensitive datasets raises serious concerns regarding privacy, security, and compliance with legal frameworks such as GDPR, HIPAA, and CCPA. Traditional centralized machine learning methods require aggregating raw data in a single repository, which poses significant risks of data breaches and unauthorized access. To address this challenge, this paper proposes a novel hybrid framework combining Federated Learning (FL) and Differential Privacy (DP) for privacy-preserving predictive analytics. The framework leverages the decentralized training capability of FL to ensure data remains localized at client devices or organizational silos, while DP mechanisms are employed to protect gradients and model updates from adversarial inference attacks.
The proposed system was tested on multiple benchmark datasets in healthcare, finance, and smart city applications. Results demonstrate that the hybrid FL+DP framework reduces privacy risks by more than 80% compared to traditional ML, while maintaining accuracy within a 3–7% margin of non-private federated models. Furthermore, the system resists gradient inversion, membership inference, and model extraction attacks, making it robust against advanced privacy threats. This research highlights the practical potential of hybrid privacy-preserving AI, setting the stage for scalable deployment in critical real-world applications.