An adaptive framework for privacy-preserving analytics in federated intrusion detection

Abstract

In this paper, we present an adaptive and energy-efficient differentially private federated learning model for IDS in IoT/IIoT environments. The method combines the Fisher Information Matrix (FIM) based parameter pruning, dynamic privacy parameter scheduling and Fast Fourier Transform (FFT) based privacy budget calculation, that strikes a balance between the data utility and the level of achieved differential privacy during training while keeping model complexity and computational overhead to minimum. The efficacy of the proposed framework is substantiated on Edge-IIoTset, a real-world dataset comprising heterogeneous and multiclass attack scenarios, across centralized and federated configurations under different client counts and simulation settings. Results show 65%–72% parameter pruning at $>$95% average accuracy over all attack types, consistent cumulative privacy budgets for varying $ϵ$-values and better generalization to non-IID data through an adaptive client selection. This manner provides a scalable privacy IDPS for edge-environment with the limited resources.