A GA-Based Feature Extraction and Its Application

In order to obtain an explicit and non-linear regress function, a new feature extraction was presented on the basis of linear support vector regression and genetic algorithm. Firstly, the linear input space in training data was mapped to a polynomial space, which can solve non-linear regression questions without complex and vague kernel skills. Then, a genetic algorithm was used to extract features from high dimension polynomial space. Suitable fitness function guaranteed that the extracted features had the biggest influence on the output in training data. Finally, linear support vector regression was introduced to the extracted features. An explicit non-linear regress function can be find. An application showed the efficiency of the new feature extraction.