Effective Sigma Selection for the Gaussian RBF Kernel

A effective method for hyper-parameter selection of the Gaussian radial basis function kernel in the one-class classification problem.

The kernel parameter σ is sensitive to the one-class classification model with the Gaussian RBF Kernel. This sigma selection method uses a line search with an state-of-the-art objective function to find the optimal value. The kernel matrix is the bridge between σ and the model. Instead of directly finding a good model, we aim to design a kernel matrix with preferred characteristics, which are reflected in the state-of-the-art objective function.

References:
[1] Zhiliang Liu*, Jinlong Kang, Xuejun Zhao, Ming J Zuo, Yong Qin, and Limin Jia. Modelling of the safe region based on support vector data description for health assessment of wheelset bearings. Applied Mathematical Modelling, 73: 19-39, 2019.