An Enhanced Privacy-Preserving Federated Learning Framework Using Attribute-Based Encryption for Smart IoT Systems

With the initiation of disruptive technologies like Internet of Things (IoT), Artificial Intelligence (AI), 5G/6G wireless communication, data plays a significant factor as an economic asset eencompassing all aspects of existence. With maximising number of data resources, privacy breaches and information leakage have become the major bottleneck in terms of maintaining the secured data exchange that creates the major threat in every human’s life. Collaborative earning is considered to be secured learning framework which is deployed for the prevention of data leakage by allowing the user’s intelligent applications to run locally. But still there exist the daunting challenges in Federated Learning (FL) frameworks when applied for the IoT based Smart Systems such as mode of sharing with designing the satisfactory incentive mechanism. In this context, novel FL framework which employs the attribute based encryption (ABE) with the Heterogeneous Chaotic Layered Encryption (HCLE) schemes that provides the fine-grained access control and assures more secure data management for the secured information sharing process in IoT based systems. The proposed FL framework encrypts the intelligent model updates through ABE-HCLE schemes, ensuring more privacy under the multiple attacks and in fully dishonest environment. The complete environment was simulated using Python 3.19 using TensorFlow-FATE FL and Charm-Crypto Libraries to deploy the proposed model in the IoT environment. The comprehensive experimentation has been conducted using BoTNET datasets thereby analysing the performance of the proposed framework in leveraging the numerous threats. Extensive simulation outcomes depict the resilience of the recommended approach to varied assaults, attaining over the 95% convergence in the privacy elevated FL rounds within 70 Interaction phases and offering solid privacy measures. Using the Optimized Light Weight Learning Model (OPLWLM), to Categorise the multiple attacks achieving the average performance of 98.5% on larger datasets. ABE-HCLE relied encryption protects the weight attributes and prevents the reliable data leakages while reducing the computational cost to 0.010 seconds every round. Theoretical and empirical outcomes assure the approach’s strength to elevate the privacy and offer the impressive operation in mitigating the multiple attacks in the smart IoT systems.