Effect of Feature Selection on Breast Cancer Data Classification using Neural Networks

Abstract

Over the past few decades, extensive death of women due to breast cancer has precipitated the need for early detection of breast cancer. Early detection of breast cancer can help in taking right decisions for the appropriate treatment plan. Breast cancer classification has become vital task for researchers and scientists. Artificial Neural Networks have been widely used in the classification task. The paper uses Multilayer Perceptron Neural Network for classifying benign and malignant tumors. One of the major challenges of classification is classifier accuracy, which can be improved using Feature Selection. Feature Selection is a strategy that aims at making classifiers more accurate and efficient by determining optimal feature subset among full set of features. The paper focuses on effect of feature selection strategies such as Correlation based Feature selection (CFS), Relief-F algorithm and Wrapper methods such as Genetic search (GA), Particle Swarm Optimization (PSO), Best First Search (BFS) and Linear Forward Selection on Breast cancer classification. The results show increased classification accuracy after applying Feature Selection. Genetic search gave more promising results in terms of accuracy as well as time required for classification compared to other strategies. Correlation coefficient, Mean Absolute Error, Root Mean Squared error, Relative Absolute Error and Root Relative Squared Error are the evaluation metrics used.