Abstract:

In order to improve the classification effect of the varied and small sample data faced by the intrusion detection field, this paper adopts the federated learning mechanism, which is widely used in distributed training recently, to solve the problem that network data is stored in independent devices and not shared with each other. This paper proposes a federated learning mechanism that integrates blockchain, which replace the central server to optimize federated learning, and designs a network intrusion detection algorithm for lightweight devices with this learning mechanism. By integrating the blockchain mechanism into federated learning, it overcomes the shortcoming of federated learning that is too dependent on a single server so as to solve the single point failure of the federated learning servers. Tested on representative data sets, the accuracy rate can reach 98.8%; In the network intrusion detection framework, the support vector machine optimized by the sparrow search algorithm is introduced. Compared with the traditional support vector machine algorithm, the accuracy rate is increased by 5.01% on average, and the false positive rate is reduced by 6.24% on average.

Key words: intrusion detection system, blockchain, federated learning, support vector machine, auto encoder