Weakly Supervised Traffic Identification Method based on Fast Detection of Angle Outliers

Current information and communication systems are facing network attacks with strong unknown characteristics, and existing network attack detection methods often fail to effectively detect unknown attacks. To address this limitation, this paper develops a novel weakly supervised traffic detection method named CTFABOD, designed to enhance the detection of unknown network attacks in contemporary information and communication systems. CTFABOD method uses generative adversarial network to enhance the performance of fast detection of angle outliers, to achieve precise detection of unknown network attacks. This paper tests the CTFABOD model on the classical NSL-KDD dataset, selects four classic weakly supervised models as comparison models, and uses AUC and precision as the evaluation indicators. The experimental results show that CTFABOD method has achieved the highest AUC and precision scores on the NSL-KDD dataset, and increased the precision score by 10.31%. This study highlights the effectiveness of combining angle-based outlier detection with generative adversarial network in improving the detection of unknown network attacks and suggests promising applications in various high-dimensional data analysis tasks.