In the rapidly growing Internet of Things (IoT) landscape, federated learning (FL) plays a crucial role in enhancing the performance of heterogeneous edge computing environments due to its scalability, robustness, and low energy consumption. However, one of the major challenges in such environments is the efficient selection of edge nodes and the optimization of resource allocation, especially in dynamic and resource-constrained settings. To address this, we propose a novel architecture called Multi-Edge Clustered and Edge AI Heterogeneous Federated Learning (MEC-AI HetFL), which leverages multi-edge clustering and AI-driven node communication. This architecture enables edge AI nodes to collaborate, dynamically selecting significant nodes and optimizing global learning tasks with low complexity. Compared to existing solutions like EdgeFed, FedSA, FedMP, and H-DDPG, MEC-AI HetFL improves resource allocation, quality score, and learning accuracy, offering up to 5 times better performance in heterogeneous and distributed environments. The solution is validated through simulations and network traffic tests, demonstrating its ability to address the key challenges in IoT edge computing deployments.

In a typical 10G-Passive Optical Network (XG-PON), the propagation delay between the Optical Network Unit (ONU) and Optical Line Terminal (OLT) is about 0.3 ms. With a frame size of 125 μs, this amounts to three frames of data in the upstream and three frames of data in the downstream. Assuming no processing delays, the grants for any bandwidth requests reach the ONU after six frames in this request-grant cycle. Often, during this six-frame delay, the queue situation is changed drastically, as much, more data would arrive in the queue. As a result, the queued data that is delayed loses its significance due to its real-time nature. Unfortunately, almost all dynamic bandwidth allocation (DBA) algorithms follow this request-grant cycle and hence lacking in their performance. This paper introduces a novel approach for bandwidth allocation, called Demand Forecasting DBA (DF-DBA), which predicts ONU’s future demands by statistical modelling of the demand patterns and tends to fulfil the predicted demands just in time, which results in reduced delay. Simulation results indicate that the proposed technique out-performs previous DBAs, such as GigaPON access network (GIANT) and round robin (RR) employing the request-grant cycle in terms of Throughput and Packet delivery ratio (PDR). Circular buffers are introduced in statistical predictions, which produce the least delay for this novel DF-DBA. This paper, hence, opens up a new horizon of research in which researchers may come up with better statistical models to brew better and better results for Passive optical networks.

As machine learning models are increasingly applied to real-world scenarios, it is essential to consider the possibility of changes in the data distribution over time. Concept drift detection and adaptation refers to the process of identifying and tracking these changes and updating the model accordingly. Researchers have devoted significant efforts to develop various techniques and tools for concept drift detection and adaptation, as this paper provides a generic roadmap and review of the field. In this paper, we begin by reviewing the background of data stream classification and its assumptions and requirements. Then, we explore the historical development of concept drift detection and adaptation and highlight the key points of approaches that have emerged over time. Next, we summarize the major findings, challenges, and limitations of past research, and provide insights into potential future directions of the field. The paper can benefit researchers and practitioners who seek to navigate the challenges and opportunities in concept drift detection and adaptation.