An Evolutionary Wrapper based Feature Selection using Modified Fire Fly Algorithm

Though breast cancer occurrence rates have increased over the years, breast cancer mortality among women of all ages has declined. Mammography has proven a reliable method to detect breast cancer at its earliest and most treatable stage. Feature selection is important to classifying images. This study presents a mammogram classification method where Gabor filters extract features and a fire fly algorithm is proposed for feature selection method. Correlation based Feature Selection (CFS) selects the initial fire flies. Classifiers like Decision Stump, J48 and CART are used for experimental result comparison. The proposed methods were evaluated by the Mammography Image Analysis Society (MIAS) database. The new feature selection achieved better result for all the above classifiers.