This investigation explores the application of combined learning for privacy-preserving machine learning in IoT systems, centring on four key calculations: Federated Averaging (FedAvg), Homomorphic Encryption-based Federated Learning, Secure Aggregation, and Differential Privacy in Combined Learning. Broad tests were conducted to assess these calculations in terms of demonstrating precision, protection conservation, and computational effectiveness. The results grandstand the taking after discoveries: FedAvg accomplished the most elevated accuracy at 92.5%, whereas Secure Conglomeration illustrated competitive precision levels at 91.8%. Homomorphic Encryption and Differential Privacy calculations showcased vigorous security conservation with negligible data spillage and security parameters of 2.5 and 1.0, separately. Secure Aggregation rose as a promising arrangement, adjusting precision and protection conservation, with negligible communication overhead. The computational productivity measurements uncovered that Secure Accumulation, in spite of its high-security conservation, caused moo communication overhead, making it appropriate for resource-constrained IoT situations. This investigation contributes to the progressing talk on combined learning in IoT, giving experiences into the trade-offs among exactness, protection, and effectiveness, and serving as an establishment for future progressions in privacy-preserving machine learning standards.

Privacy Preservation, Federated Learning, Machine Learning Algorithms, IoT Networks, Computational Efficiency