Parameter Selection of Support Vector Machines for Cooperative Spectrum Sensing Against Probabilistic Byzantine Attack in Cognitive Wireless Sensor Networks

Cooperative spectrum sensing (CSS) enables secondary sensor nodes in cognitive wireless sensor networks (CWSNs) to cooperatively detect spectrum holes while safeguarding primary users (PU) from interference. Unfortunately, Byzantine attack pose a significant threat to the reliability of CSS. To further distinguish normal sensor nodes (NSN) from malicious sensor nodes (MSN) and mitigate the negative impact of Byzantine attack on CSS, this letter introduces support vector machine (SVM) as a solution because the efficacy of SVMs hinges on the proper selection of numerous parameters, including the kernel function and regularization functions. Initially, we improve the Byzantine defense reputation algorithm by incorporating a historical attack count accumulation mechanism and SVM, and designs a series of SVM parameter combinations to optimize performance. Furthermore, we train an SVM decision algorithm based on real-time reputation to acquire and weigh nodes' historical sensing reputation, which significantly enhances the algorithm's accuracy. Subsequently, we design and experiment with a range of SVM parameter combinations to identify the optimal solution. Finally, numerical results demonstrate substantial improvements in the system reliability, validating the effectiveness of our proposed CSS algorithm.