Daniel C. Moura, Miguel A. Guevara López, Raul Ramos Pollan, Isabel M.A. Pereira Ramos, Joana Pinheiro Loureiro, Teresa Cardoso Fernandes, Bruno M. Ferreira de Araújo
ICEM15: 15th International Conference onExperimental Mechanics, FEUP-EURASEM-APAET,
Porto/Portugal, 22-27 July 2012
This work compares the performance of different supervised learning algorithms for the task of classifying breast lesions as suspicious or not of malignancy. Three algorithms were compared: Linear Discriminant Analysis (LDA), Support Vector Machines (SVM), and Random Forests (RF). Experiments were performed on the Portuguese Breast Cancer Digital Repository (BCDR).
Results show that choosing the best classifier is a problem dependent of the amount of training data available, with SVM being the classifier converging faster and therefore best on smaller datasets, while RF being best on larger datasets.