A Federated-ANFIS for Collaborative Intrusion Detection in Securing Decentralized Autonomous Organizations
提出一种联邦自适应神经模糊推理系统（FANFIS），用于区块链-物联网网络中的协作入侵检测，在保护隐私的同时将模型训练时间平均减少49.42%，且性能优于传统集中式方法。
Blockchain has facilitated the emergence of automation and decentralization concepts, leading to significant organizational and operational changes in businesses, e.g., decentralized autonomous organizations (DAOs). In DAOs, management decisions are made collectively and automatically through smart contracts without a central authority, which results in increased cybersecurity requirements. While blockchain integration aims to eliminate single points of failure and enhance data integrity, DAOs remain susceptible to vulnerabilities in consensus mechanisms, key management, and software management, highlighting the need for intrusion detection. Collaborative intrusion detection has been identified as a potential solution to address emerging cyberattacks in a decentralized environment; however, it is not yet fully developed. This study proposes a federated adaptive neuro-fuzzy inference system (FANFIS) for collaborative intrusion detection in blockchain–Internet-of-Things (IoT) networks. The FANFIS maintains a global intrusion detection model in a privacy-preserving manner over the network. Through computational experiments with datasets of KDDCUP99 and Bot-IoT, we found that using the FANFIS reduced the computational time for model training by an average of 49.42% while maintaining a high-performance level. The superior performance of the FANFIS, as demonstrated by its accuracy, precision, and F1-score, surpasses the conventional method involving data centralization, exhibiting mean percentage errors of 1.4092%, 2.6935%, and 1.3463%, respectively.