Enhancing Clustered Federated Learning Using Artificial Bee Colony Optimization Algorithm for Consumer IoT Devices

Consumer Internet of Things (CIoT) interconnects multiple devices over Internet, like smartphones, wearables, and smart gadgets to simplify tasks and provide convenience. However, it encounters obstacles such as privacy apprehensions arising from data aggregation, security flaws, interoperability discrepancies. Federated learning (FL) mitigates these issues by localizing data, reducing privacy risks, and securing IoT networks. It stores and updates learnt models on a central server, which is necessary for CIoT networks. Nonetheless, its application confronts problems like non-independent and identically distributed (non-IID) data, communication efficiency, and privacy issues. According to recent research, training models using non-IID data have detrimental influence on performance, convergence, and overall model quality in FL. Moreover, traditional FL approaches, including clustered federated learning (CFL), have problems with client training and fixed hyperparameter use. This paper introduces new approach, Artificial Bee Colony Clustered Federated Learning (ABC-CFL). ABC-CFL uses Density-based spatial clustering of applications with noise (DBSCAN) to cluster client devices based on training hyperparameters, followed by hyperparameter optimization for each cluster to better suit each cluster using artificial bee colony algorithm. This method outperforms static hyperparameter utilization problems and improves model performance and communication efficiency in CFL as demonstrated by experiments on the CIFAR-10, MNIST and CelebA datasets.