Breast Cancer Classification by Using Support Vector Machines with Reduced Dimension

Abstract

Correct and timely diagnosis of diseases is an essential matter in medical field. Limited human capability and limitations decrease the rate of correct diagnosis. Machine learning algorithms such as support vector machine (SVM) can help physicians to diagnose more correctly. In this study, Wisconsin diagnostic breast cancer (WDBC) data set is used to classify tumors as benign and malignant. Independent component analysis (ICA) is used to reduce the dimensionality of WDBC data into two feature vectors. The effect of using two reduced features to classify breast cancer with SVM and polynomial or radial basis function (RBF) kernels are investigated. Performances of these classifiers are evaluated to find out accuracy, sensitivity and specificity. In addition, the receiver operating characteristics (ROC) curves of SVM with these kernels are presented. Results show that SVM with quadratic kernel provides the most accurate diagnosis results (94.40%) and decreases the accuracy and sensitivity values slightly when the dimensionality is reduced into two feature vector computing two independent components.