Dynamic Communication Network Anomaly Detection Based on Interactive Learning
In dynamic communication networks, issues such as traffic distribution drift can lead to a decline in the generalization ability of detection models, while high-dimensional feature redundancy and other factors limit the efficiency of model updates. To address these issues, this paper proposes UCB-HADDN, a dynamic anomaly detection method based on interactive learning, aimed at achieving efficient anomaly detection within a limited labeling budget. The method employs a multi-clustering feature selection algorithm for unsupervised feature filtering and constructs an interactive detection framework that combines clustering structures with communication context. By introducing a confidence upper bound strategy, the method achieves continuous adaptation to changes in traffic distribution. We conducted experiments on three typical dynamic scenarios using the CICIDS2017 dataset. The results show that, in the distribution-aligned scenario, UCB-HADDN achieved a precision of 0.96, with a query efficiency 2.90 times that of the random strategy; In the novel-attack scenario, UCB-HADDN achieved a query precision of 0.57 during the frozen-parameter test phase, outperforming the comparison baselines. After enabling online interactive updates in the new phase, query precision increased to 0.94, while anomaly coverage increased from 0.23 to 0.37 under a fixed budget of 300 queries. The proposed method demonstrates high query efficiency and strong adaptability to dynamic traffic conditions under the evaluated experimental settings. Additional runtime experiments indicate relatively low per-query execution time and moderate memory consumption. These results suggest the computational feasibility of the proposed interactive detection framework.
Authors
- Longwen Wu (ORCID: https://orcid.org/0000-0002-6914-6695)
- Dongming Luo
- Yaqin Zhao (ORCID: https://orcid.org/0000-0002-0167-0597)
- Shengyang He (ORCID: https://orcid.org/0000-0003-1301-3524)
- Xiangrui Hu (ORCID: https://orcid.org/0009-0009-6633-4111)
- Lei Li
Institutions
- Harbin Institute of Technology (CN)
- China Electronics Technology Group Corporation (CN)
Publication Details
- Journal
- Electronics
- Published
- 2026-11-13
- DOI
- https://doi.org/10.3390/electronics15225288
- Primary Topic
- Network Security and Intrusion Detection
- Type
- article
- Field-Weighted Citation Impact
- 0.00
Funders
- National Natural Science Foundation of China