Fed-FIDS: A Efficient Federated Learning-based Intrusion Detection Framework

Intrusion detection systems are essential for identifying and mitigating threats in rapidly growing and complex networks, but the centralized processing of sensitive data raises privacy concerns. Federated learning (FL) has emerged as a promising approach for collaboratively training intrusion detection models while preserving data privacy by keeping data locally on edge devices. However, the heterogeneous computational capabilities of edge devices and varying data quality among clients pose significant challenges to the efficiency and accuracy of federated learning-based intrusion detection systems. To address these issues, we propose Fed-FIDS, an efficient federated learning framework for intrusion detection. Fed-FIDS incorporates three key modules: Real-time Contribution Quantification, Resource Allocation, and Data Selection, which work together to optimize the allocation of computing resources and the selection of local training data. Experimental results on UNSW-NB15 and Edge-IIoT datasets using DNN, 1D-CNN, and 1DCNN-LSTM models demonstrate that Fed-FIDS consistently outperforms baseline approaches, improving attack detection accuracy by up to 4.98% while accelerating the training process by an average of 144 times. Fed-FIDS's resource allocation mechanism is validated through comparisons with random and equal allocation strategies, showcasing its reliability in rewarding clients with high-quality data and sufficient computational capabilities. Our findings highlight the potential of Fed-FIDS as an efficient and accurate solution for intrusion detection in edge computing environments.