Effects of large constituent size in variable neural ensemble classifier for breast mass classification

Abstract

This paper proposes a novel ensemble technique for mass classification in digital mammograms by varying the number of hidden units to create diverse candidates. The effects of adding more networks to the ensemble are evaluated on a mammographic database and the results are presented. A classification accuracy of ninety nine percent is achieved. © Springer-Verlag 2013.