A Random Feature Selection Method for Classification of Mammogram Images

Abstract

This article discusses the use of a random feature selection method for classification of mammogram images using a multi-scale transform. Each image is represented by a vector of coefficients. Subsets of columns are randomly generated and used for classification of a training set. The subsets achieving a predefined performance are kept and pooled in a final set for testing. The method is tested using a set of images provided by the Mammography Image Analysis Society (MIAS) to differentiate normal and abnormal images. In our experiments the classifiers K nearest neighbors (kNN) and Discriminant Analysis (DA) are used with Wavelet transform.