IoT Revolutionization using Federated Learning and Decentralized AI on Edge Devices

This research proposes a novel strategy for addressing the limitations of centralized architectures in IoT data processing. Traditional systems experience significant bandwidth use, privacy difficulties, and scalability issues. To address these difficulties, the proposed approach combines Federated Learning (FL) with Decentralized AI on Edge Devices (DAI-ED). By decentralizing processing duties and utilizing FL methods, data is kept local to edge devices, lowering latency and improving privacy. Deployment entails choosing AI-capable edge devices and configuring them for seamless integration. Secure communication techniques provide privacy throughout FL operations. Model optimization approaches help to improve efficiency even further. Compared to centralized systems, the results show higher model accuracy, decreased communication overhead, and increased resource utilization. The proposed system has an overall accuracy of $\mathbf{9 5 \%}$, with edge device accuracy at $\mathbf{9 4. 5 \%}$ and central server accuracy at $\mathbf{9 6 \%}$. Communication overhead during the training and inference phases is greatly minimized, with the proposed system requiring 150 MB per device against $\mathbf{4 0 0} \mathrm{MB}$ in centralized systems. The proposed system has low memory consumption and computational complexity, but it is very scalable.