A Novel Method of Intrusion Detection Based on Federated Transfer Learning and Convolutional Neural Network
Xiang Ji, Hong Zhang, Xulun Ma · 2022 IEEE 10th Joint International Information Technology and Artificial Intelligence Conference (ITAIC) · 2022
As a network security defense technology, intrusion detection system can effectively protect network security. At present, machine learning is widely used in intrusion detection and has achieved good application results. The detection methods based on traditional machine learning need enough available intrusion detection data samples, and the samples meet the conditions of independent and identically distributed. However, in reality, the intrusion detection data generated by a single institution is insufficient, and various institutions protect users' privacy and data security in the form of islands, which makes it difficult to maintain the same data distribution. In addition, there is the problem of data imbalance. To solve the above problems, this paper proposes a new intrusion detection method FTLCNN, which integrates federal transfer learning and convolutional neural network. FTLCNN constructs a transfer convolution neural network framework to solve the problems of sample scarcity, imbalance and probability adaptation; under the mechanism of federal learning, FTLCNN use the model to learn without sharing training data, protect data privacy and solve the problem of data island. The experimental on UNSW-NB15 shows that compared with the other four benchmark algorithms, FTLCNN has higher detection rate and lower false positive rate, and has significant advantages in solving the problem of scarcity and imbalance of in intrusion detection.