Diagnosing breast masses in digital mammography using feature selection and ensemble methods

Methods that can accurately predict breast cancer are greatly needed and good prediction techniques can help to predict breast cancer more accurately. In this study, we used two feature selection methods, forward selection (FS) and backward selection (BS), to remove irrelevant features for improving the results of breast cancer prediction. The results show that feature reduction is useful for improving the predictive accuracy and density is irrelevant feature in the dataset where the data had been identified on full field digital mammograms collected at the Institute of Radiology of the University of Erlangen-Nuremberg between 2003 and 2006. In addition, decision tree (DT), support vector machine-sequential minimal optimization (SVM-SMO) and their ensembles were applied to solve the breast cancer diagnostic problem in an attempt to predict results with better performance. The results demonstrate that ensemble classifiers are more accurate than a single classifier.