A sustainable and trusted solution for IoT-based federated learning using feedback behavior

Abstract

Federated learning is defined as one of the solutions that use distributed clients to train and aggregate the target model without sharing the private information of clients in the network. Hence, federated learning may easily handle and tackle the generated information in a more convenient, feasible and secure manner, while involving the intelligent devices to perform testing and training at the local edges. However, the design of such federated learning solutions faces multiple difficulties due to the cost and complexity of information computation, along with a high energy consumption. Furthermore, the identification of resource-constrained devices which ensures significant power consumption along with privacy and security to the devices makes it critical to attain a sustainable federated learning environment. The aim of this paper is to integrate feedback-based and objective trust models to accurately identify the behavior of communicating devices in the network. The secure transmission of information forwarded by trusted IoT devices is computed by the objective model. In addition, the energy consumption required for measuring the authenticity of a device can be further diminished using a behavior and feedback trusted model. In addition, the past history behavior and feedback received from neighboring devices may also contribute to achieving an efficient and sustainable security mechanism with reduced communication steps and computational delay in the network. The proposed solution is evaluated to analyze the feasibility and performance in terms of energy consumption and resource utilization, while also considering various security metrics for comparison against existing methods. The proposed mechanism out-performed 87% improvement in terms of security metrics in comparison of existing approaches.