A Robust Blockchain Security Framework for IoT Using Federated Learning-Based Anomaly Detection

The increase of botnet attacks and blockchain anomalies in decentralized networks especially in the case of the IoT environments requires strong detection mechanisms that preserve privacy. This research provides an innovative solution in the form of Dynamic Weighted Updating Federated Averaging (DWUFedAvg) that integrates federated learning with blockchain in order to improve irregularity recognition in IoT devices. As compared to conventional FedAvg, DWUFedAvg adjusts model weights dynamically, leading to the reduction of the influence of underperforming nodes. Using blockchain, the model makes sure model updates remain transparent, auditable and tamper-proof aggregated. The approach was tested with BotIoT and NBaIoT datasets, which showed better performance in terms of metrics as compared to previous approaches. Not only does the integration of blockchain ensure that model updates are secured, but is also in line with the decentralized nature of federated learning, whilst maintaining data privacy and latency reduction. The outcome reveals how DWUFedAvg is capable of detecting botnet attacks effectively and with an ability to take care of scalability, transparency and security issues that have been forth-coming in federated settings. This research informs the bases of secure, scalable and efficient anomaly detection models, especially for decentralized and data-sensitive systems, and advises areas of advancing existing and new models with regards to dynamic weighting, blockchain scalability and model fairness. The suggested DWUFedAvg technique achieves an overall accuracy of 97.55%, surpassing BlockHunter (90.0%), FedGroup (88.45%), and FedAvg (86.35%) on the BotIoT and N-BaIoT datasets.