Performance Analysis of Breast Cancer Classification Using Feature Selection and Machine Learning

Abstract

Breast cancer remains one of the leading causes of cancer-related deaths in women globally and, therefore any means of diagnosing this disease, accurately and early enough is quite important in handling it. Here in this paper, author aims to compare the result of different machine learning algorithms in classification of breast cancer based on feature selection. In the present study we utilize Dataset that includes clinical and imaging data derived from breast cancer patients. Classification models are regularized with Recursive Feature Elimination (RFE) and Principal Component Analysis (PCA) in an attempt to manage and minimize the dimensionality of certain datasets. Some of the learning algorithms that are used include SVM, Random forest and the k-NN algorithms with the selected features being used to train and test the various algorithms. Classification models’ efficacy is measured by performance indicators: accuracy, sensitivity, specificity, and AUC-ROC. In our experiments, we have observed an area under the curve and reduced classification error along with increased computation time and iter – a consequence of selection or rejection of features. Moreover, we compute the discriminant features detectable to classify breast cancer well. The knowledge gained from this experience can be of significant help in the improvement of current machine learning methods aimed at diagnosing breast cancer, and may help to identify this illness at an early stage and significantly enhance people’s quality of life.