基于再生核希尔伯特空间映射的高维数据特征选择优化算法.

The existing filler feature selection algorithms do not consider the inner structure of nonlinear data, lead to a lower classification accuracy than wrapper feature selection methods. This paper proposed a reproducing kernel Hilbert space mapping based feature selection algorithm to solve that shortcoming of filter feature selection algorithms. Firstly, it constructed the search tree based on branch and bound method and searched. Then, based on the reproducing kernel Hilbert space mapping, it analyzed the inner structure of nonlinear data. Lastly, based on the inner structure of the data, it selected the optimal distance computing method. Compared simulation experiments results show that the proposal has a similar classification accuracy with wrapper feature selection algorithms, at the same time has obviously heller computational efficiency, and can handle the big data analysis. [ABSTRACT FROM AUTHOR]