Comparative Analysis of ANN and SVM in Digital Mammogram based on Feature Selection

Abstract

This research paper focuses on feature extraction and selection strategies for detecting and classifying malignant tumors in mammograms. Features, the distinctiveness of the objects of attention, if chosen carefully, are envoy of the greatest relevant information so that the image has to suggest for absolute characterization of an abrasion. Feature extraction techniques analyze the images of an object to extract the most important features that are representative characteristics of the various classes of objects. Features are used as inputs to distinguish that allocate them to the class that they signify. GLCM (Gray Level Co-occurrence Matrix) is the universally employed technique for texture examination and it compares the gray-level divergence of any two neighboring pixels in a specified displacement and direction on an image. GLCM of an image encompasses a function of the angular connection and an interval between pixels in the neighborhood. In this paper GLCM method has been used to extract features in the proposed CAD system. Feature selection is a technique generally employed for data mining and knowledge exploration to minimize dimensionality and allows the removal of redundant features, keeping the essential hidden information simultaneously, and selection of features requires a reduced amount of data show and efficient data mining. In respect of packet collisions, data rate, and storage, it also brings potential communication benefits. Feature selection is a significant key point in machine learning and other related medical fields. In this paper, the proposed feature selection method is employed with existing classification techniques for the results, before and after feature selection. Support Vector Machine (SVM) and Artificial Neural Network (ANN) classification algorithms were used for validation. The findings show that the suggested strategy is capable of outperforming existing feature selection techniques in terms of classification performance.