Federated MapReduce fractional deep learning for scalable and privacy-aware intrusion detection in cloud computing

Conventional intrusion detection systems (IDSs) are not suitable for large-scale, heterogeneous, and privacy-sensitive cloud environments, owing to the rapid adoption of cloud computing and Internet of Things (IoT) technologies, which have resulted in an increasing scale and complexity of cyberattacks. Current machine learning, deep learning, and federated IDS methods are typically centralised, cannot be scaled to provide high service availability, are inefficient at distributing data preprocessing, are not robust enough to converge under non-IID data distributions, and lack proper adaptive optimisations. In order to overcome these drawbacks, in this paper, a Federated MapReduce-Enabled Fractional Deep Learning Framework (FedMR-FFNet) is proposed to provide scalable and privacy-preserving intrusion detection in cloud computing. The proposed structure includes MapReduce-based distributed preprocessing to process features in a large scale efficiently, HSBOA-driven ensemble feature selection to select informative features of an intrusion, Fractional Feedforward Network (FFNet) learning to address the issues of complex nonlinear intrusion pattern representations, Taylor Wave Search Algorithm (TWSA) for optimizing the model parameters adaptively, and an adaptive federated aggregation strategy to assign weights to the clients based on their dataset size, loss, update consistency, and noise level to enhance the robustness under heterogeneous non-IID environments. The proposed framework is validated through extensive experiments conducted on the NSL-KDD, CIC-IDS2017 and TON_IoT datasets. The methods of FedMR-FFNet show high accuracy (99.18%), precision (99.03%), recall (98.94%) and F1-score (98.98%), and the convergence speed is stable, reducing the computational time and improving the accuracy, precision, recall and F1-score relative to conventional machine learning, deep learning and federated IDS. Moreover, the framework exhibits high cross-dataset generalisation, communication efficiency, scalability and privacy preservation, which presents a promising solution to intelligent intrusion detection in modern distributed cloud and IoT infrastructures.