| Title |
SVM-DBSCAN Dual Detection Structure for Adversarial Jamming Detection and Defense |
| Authors |
박예은(Ye-Eun Park) ; 김혜리(Haeri Kim) ; 정종문(Jong-Moon Chung) |
| DOI |
https://doi.org/10.5573/ieie.2025.62.9.30 |
| Keywords |
Jamming detection; Adversarial defense; SVM; DBSCAN; Smart city networks |
| Abstract |
With the increasing use of drones and autonomous vehicles in smart city environments, the threat of jamming attacks, a wireless communication-based security threat, is on the rise. In particular, there is a need for AI-based hostile jamming countermeasures that degrade the performance of machine learning classifiers by subtly modifying the input data. This study proposes a dual detection structure of SVM-based jamming classification and DBSCAN -based outlier removal to effectively respond to hostile jamming attacks. First, a jamming detection model is built based on the RBF kernel of SVM. Then, among the data classified as normal by SVM, outliers near the decision boundaries are removed by DBSCAN to mitigate the detection performance degradation caused by hostile attacks. Experiments were conducted under varying attack ratios. The results show that the proposed method with SVM and DBSCAN dual detection maintains a stable detection performance even under adversarial conditions, and that performance degradation below a certain level can be used as a signal to change the encryption algorithm to reinforce the security level. |