A novel feature selection method based on data normalization

Data normalization is the first step of pattern recognition (PR). Feature selection is another part of work in PR. All of them are very important and challengeable. In this paper, some normalization methods and some feature selection methods are introduced first, then, discussed the relationship between feature selection and data normalization. Finally, an algorithm is described how to normalize supervised multi-class (phenotype) data that all the classes are the type of multi-sample by a statistical way. By the method, on the training phase of PR, useful features can be obtained with the same normalization algorithm. The method in this paper has practical significance.