Optimization of distributed network intrusion detection system based on internet of things and federated learning

Abstract The Internet of Things (IoT) has been proposed to pose a greater risk of cyberattacks due to the large amounts of data traffic and the diverse range of devices. The main limitations of traditional centralized intrusion detection systems (IDSs) are attributed to privacy risks, high communication costs, and poor scalability. The research presents a distributed, privacy-preserving framework for intrusion detection, which combines Federated Learning (FL) with a new Deep Learning model that performs and optimizes network intrusions to collect and analyze aspects of “federated” augmentation, then improve security in Web usage. The particular method includes Recursive Feature Elimination (RFE) for the reduction in characteristics, the Federated Kalman Filter (FKF) to reduce noise, and an Adaptive Artificial Fish Swarm optimized Long Short-Term Memory (AdapAFS-LSTM) model for accurate detection of multi-type network intrusions. The model parameters are distributed based on IoT model nodes and do not share raw data. Model parameters learn from IoT nodes, which are combined based on the Federated Proximal (FedProx) algorithm and can be applied toward the development of a robust global IDS. Experimental evaluation of the distributed and privacy-preserving intrusion detection framework on the Multi-Type Network Attack Detection (M-TNAD) dataset demonstrated superior performance in achieving 99.79% accuracy, F1-score, precision, and recall, showing low resource consumption in the final execution time and performance metrics. This work demonstrates the potential of implementing a federated, optimization-driven deep learning method to effectively develop an IDS solution against IoT networks through optimization methodology and machine learning.