Comparative Analysis of Feature Selection Methods for Blood Cell Recognition in Leukemia

摘要

This study analyses different methods of diagnostic feature selection in the problem of classification of the blood cells in leukemia. The analyzed methods belong to the wrapper and filter methods and cover wide range of approaches to feature selection problem. In particular they cover 7 methods, each of them working on different principle. As a results of this preprocessing stage we define the best (according to the applied method) set of features which is next used as the input for the Gaussian kernel SVM classifier. The last step of blood cell recognition is the integration of the results of application of all methods. The numerical results of experiments will be presented and analyzed.