A Wrapper Feature Selection Technique for Improving Diagnosis of Breast Cancer

Abstract

Nowadays, cancer is considered as a fairly common disease. Regarding the number of newly detected cases, breast cancer is ranked as one of the most leading cancer types to death in women. It can be cured, if it is identified and treated in its early stages. Therefore, this study explores a proposed integrated wrapper feature selection method called wrapper naïve-greedy search (WNGS) to improve the accuracy of the breast cancer diagnosis. WNGS is based on a wrapper method, which is blended with a greedy forward search to select optimal feature subset. WNGS method integrates a wrapper method based on Naïve Bayes (NB) classifier as a learning scheme with a forward greedy search method. Then, the selected feature subset is fed to a classifier to determine breast cancer. In addition, K-nearest neighbor-greedy search (KNN-GS) is used for comparison. In KNN-GS method, k-nearest neighbor (KNN) classifier is used as a learning scheme while a forward greedy search method is used to search through features. NB is used as the classifier for classification process for both methods. By applying these two methods, data features are reduced, and the classification rate is improved. Both methods are tested on two different benchmark breast cancer datasets. Accuracy results showed that WNGS method outperformed KNN-GS method. Also, WNGS method overcame KNN-GS regarding precision, recall, F-measure, and sensitivity.