Semi supervised learning methods for intrusion detection based on F CSGRU

DOI:	10.11809/bqzbgcxb2023.05.037
Keywords:	semi supervised learning; network intrusion detection; cost sensitive; fuzziness; gated recurrent unit
Abstract:	The existing supervised learning methods can only use labeled samples to train the classifier, which is difficult and costly to obtain labels. At the same time, the obtained samples are prone to sample class imbalance, which seriously affects the analysis ability of intrusion detection models. In order to solve the above problems and improve the effect of the intrusion detection model, this paper proposes a semi supervised learning method of intrusion detection based on fuzzy cost sensitive gated recurrent unit (F CSGRU). This method uses semi supervised learning and cost sensitive methods to improve the classifier performance of the intrusion detection system, and improve the detection ability of samples of less common classes. The model combines cost sensitivity with gating cycle units to generate labels for unlabeled samples, and divides the samples according to fuzzy entropy, in which low fuzzy entropy samples are merged into the original training set to train the classifier again. Based on NSL KDD and UNSW NB15 data sets, the paper conducts a comparative test. The results show that the accuracy of the model proposed in the paper to the above data sets can reach 99.30% and 84.53% respectively, and, compared with the classic CNN BiLSTM, the accuracy rates improve by 0.08% and 2.45% respectively. The improvement effect is particularly significant for the detection accuracy of samples of less common classes.
Issue:	Vol. 44 No. 5 (2023)
Published:	2023-05-28
PDF HTML