Regularized Deep Neural Network in Identification of Breast Cancer

Abstract

Breast cancer is very common and hazardous diseases amongst the women. Thousands of women die every year due to breast cancer. It is very essential to diagnose the breast cancer in its early stage. To get success in dealing with this problem, Deep Neural Network (DNN) is introduced. DNN has played a wide role in medical science in diagnosis of many diseases. In this paper, we designed regularized DNN for prediction of breast cancer. We simulated regularized DNN by employing different optimization algorithms like L-BFGS, SGD, Adam and with different activation functions like Logistic, Tanh, and ReLu. Independent Component Analysis (ICA) technique is employed for feature selection. The proposed network is trained and tested on WDBC dataset from UCI Repository of Machine Learning database and detailed analysis is carried out along with the speed of convergence of each algorithms. The network designed has achieved highest 100% accuracy for the dataset in just few seconds.