Unsupervised nonlinear feature selection algorithm via kernel function

Feature selection is one of the important methods of data preprocessing, but the general feature selection algorithm has the following shortcomings: (1) Noise and outliers cannot be ruled out so that the algorithm does not work well. (2) They only consider the linear relationship between data without considering the nonlinear relationship between data. For this reason, an unsupervised nonlinear feature selection algorithm via kernel function is proposed in this paper. First, each data feature is mapped to a kernel space by a kernel function. In this way, nonlinear feature selection can be performed. Secondly, the low-rank processing of the kernel coefficient matrix is used to eliminate the interference of noise samples. Finally, the feature selection is performed through a sparse regularization factor in the kernel space. Experimental results show that our algorithm has better results than contrast algorithms.