Using machine learning tool in classification of breast cancer

This research implements a feed forward back propagation network (FFBPN) for classification of breast cancer cases to malignant or benign. The purpose of the research is to design an Artificial Neural Network (ANN) with high and acceptable level of accuracy by selecting the number of hidden layers, number of neurons in the hidden layer and the type of activation functions in hidden layers. Samples for training and validation of ANN are obtained from Wisconsin Breast Cancer Database (WBCD) which is open access dataset. The dataset contains 699 samples that were distributed to two groups: 599 samples in training setand 100 samples in testing set. Each sample has 9 attributes representing 9 characteristics of breast fine-needle aspirates (FNAs) as inputs of the network. This experiment includes a comparison among the obtained mean square error (MSE) when using three transfer functions: LOGSIG, TANSIG, and PURELINE in neural network architetcures. Impact of different number of layers (1, 2, and 3 layers were used)in ANN architectureon output accuracy was also investigated. Also, this research provides the results of ANN performance for different number of neurons in hidden layer (20, 21, 22, 23, 24 neurons were implemented). The results show that the best network design is that one with three hidden layers, 21 neurons in the hidden layer, and TANSIG as activation function.