Performance Validation of Spectrum Sensing Using Kernelized Support Vector Machine Transformation

Due to the increasing interest in wireless networks, the availability of spectrum has become a challenge. With the help of cognitive radio, a promising technology, can be overcome this issue. One of the most challenging tasks in this technique is finding the available spectrum holes. The increasing interest in machine learning techniques for spectrum sensing (SS) has led to the development of several novel methods. In this paper, we use the support vector machine with the kernel transformation that are designed to improve the performance of SS, such as such as Linear kernel, Radial Basis Function or Gaussian kernel, Polynomial kernel and Sigmoid kernel. One of the main reasons why the kernel functions are used is due to the possibility of having a non-linear dataset. The performance of kernel functions is compared in terms of accuracy, precision, recall, f1_score and confusion matrix for different number of users such as 100, 500 and 1000. Among all these, Polynomial kernel SVM has shown better performance of 96%, 97% and 100% accuracy for 100, 500 and 1000 number of users. In addition, this paper presents a comparison of the proposed and existing methods, where the proposed method has shown a better performance.

Automated summary

The increasing interest in wireless networks has made spectrum availability a challenge. Cognitive radio, a promising technology, can help overcome this issue. Finding available spectrum holes is one of the most challenging tasks in this technique. The use of machine learning techniques, particularly support vector machine with kernel transformation, has improved spectrum sensing performance.