An Intrusion Detection Method Integrating KNN and Transfer Extreme Learning Machine
Kunpeng Wang, Jingmei Li · 2022 2nd Asia-Pacific Conference on Communications Technology and Computer Science (ACCTCS) · 2022
As a network security defence technology, an intrusion detection system (IDS) can effectively detect network attack behaviour, significantly protecting network security. Machine learning has been widely used in intrusion detection to improve network intrusion detection and shows the advantages of more intelligence and accuracy than traditional methods. However, intrusion detection based on conventional machine learning requires that the training samples meet the conditions of independent and identical distribution, and data imbalance affects modelling and training. This paper proposes a new intrusion detection model KnTrELM, which integrates KNN and transfer extreme-learning-machine to solve the problems. KnTrELM first applies KNN to detect and delete outlier data to obtain a small-scale and high-quality training data set. Then, based on the transfer learning and extreme learning machine, a transfer extreme learning machine method is proposed. It utilizes the correlation between a large number of labelled data in the source and the target domain data with only a small number of samples, through the probability adaptation between domains, transferring the similar knowledge of source to the target domain, to improve the learning ability of unbalanced samples. Compared with the other four benchmark algorithms, KnTrELM dramatically shortens the training time, increases the training efficiency, and enhances the detection accuracy.