Recommender systems and mobile crowd sensing applications rely heavily on user-generated data, raising significant concerns about data privacy, confidentiality, and misuse. PrivFusion is developed as a privacy-preserving framework that integrates federated learning, differential privacy, clustering, and blockchain-based audit mechanisms to maintain security without compromising system performance. The system ensures that sensitive user data remains on local devices, differential privacy noise protects model updates, clustering improves aggregation efficiency, and blockchain provides immutable verification of data transactions. Experimental evaluation performed using simulated healthcare data demonstrates that PrivFusion maintains strong accuracy even under strict privacy budgets, while the blockchain layer enhances transparency and traceability. The results indicate that PrivFusion is scalable, reliable, and suitable for real-world privacy-sensitive environments.