Fast Modular Artificial Neural Network for the Classification of Breast Cancer Data

Abstract

Medical data mining is both an interesting as well as a competitive field, where research is carried out to invent new algorithms and techniques which can aid in efficient classification and prediction of various diseases. Breast cancer is one among the deadly diseases killing large population of women around the world. Many techniques have been proposed for classification of breast cancer data, but Artificial Neural Network ranks first in the accuracy rating. In this paper we come up with an idea of Fast Modular Artificial Neural Network (FM---ANN), where a feature selection step followed by data normalization (range of -1 to 1), data set division and attribute division leads to the fine refinement of input making them more suitable for classification. Modular Neural Network is built using four different types of Feed forward Neural Network (FNN) and refined inputs are sent to each module where they carry out their task distinctly. The final result of the model is the probabilistic sum of results obtained from all modules. The FM--ANN produces highest classification results compared to other networks. We tested our model on two different benchmark data sets and obtained good results. The accuracy of Wisconsin Breast Cancer Diagnostic Data (WBCD) is found to be 99.8% and, accuracy of KDD cup 2008 breast cancer data is found to be 99.96%.