Machine learning techniques for classification of breast cancer

Abstract

The major challenge in cancer diagnosis is the number of patients who are incorrectly diagnosed. To address this, we have developed and tested different expert diagnostic systems which differentiate among patients with and without breast cancer based on samples describing characteristics of the cell nuclei present in the digitized image of a fine needle aspirate (FNA). Data was collected from the UCI machine learning respiratory, specifically 699 samples. Our results demonstrate that a Feed Forward Backpropagation single hidden layer neural network with 20 neurons and TANSIG transfer function has the highest classification accuracy (98.9% and 99% accuracy in training and test set, respectively). The accuracy of multilayer architectures was significantly lower, and valued between a range of 74.9–86.3%, where the average was 81.37%. A developed expert system with a proven accuracy can be used in the future in laboratory conditions as a promising method for early classification diagnosis for breast cancer.