Breast Cancer Diagnosis Using Genetic Programming Generated Feature

This paper proposes a novel method for breast cancer diagnosis using the feature generated by genetic programming (GP) based on Fisher criterion. GP as an evolutionary mechanism provides a training structure to generate features. Fisher criterion is employed to help GP optimize features whose values corresponding to pattern vector belonging to the same class are extremely similar while those corresponding to pattern vectors belonging to different classes appear very different. The presented approach is experimentally compared with some classical feature extraction methods. Results demonstrate the capability of this method to transform information from high dimensional feature space into one dimensionality space and automatically discover the relationships among data, in order to improve classification accuracy