Comprehensive Fairness Analysis in Federated Learning with Green Anaconda Optimization: Evaluating Fair Performance in IoT Environments

Federated Learning (FL) enables the distributed training of AI models on client devices while preserving data privacy. Client fairness remains a critical challenge in FL, especially in heterogeneous Internet of Things (IoT) environments with non-Independent and Identically Distributed (non-IID) data. This paper presents a comprehensive fairness analysis of the Federated Green Anaconda Optimizer (FedGAO) method. Through extensive experiments on CIFAR-۱۰ with non-IID distribution, we evaluate multiple fairness metrics, including Jain Index, Gini Coefficient, accuracy disparity, and statistical parity. Results show FedGAO achieves near-perfect fairness (Jain Index: ۱.۰۰۰۰) and reduces accuracy disparity from ۱۰.۷۸% to ۱.۱۶% (۸۹% improvement). The method attains ۹۹.۵۷% mean client accuracy, demonstrating uniform performance across heterogeneous clients. Additionally, FedGAO achieves ۱۲.۵۰% higher global accuracy and ۸۰% lower communication costs than FedAvg. These findings establish FedGAO as an efficient approach that excels in accuracy, communication efficiency, and client fairness for Edge-IoT applications.

کلیدواژه ها:

Federated Learning ، Client Fairness ، Green Anaconda Optimization ، Metaheuristic Algorithm ، Internet of Things ، Non-IID Data