최적의 서포트 벡터 머신을 이용한 유방암 분류

Abstract

Objectives: This paper is to examine breast cancer classification using support vector machine (SVM). SVM with optimal parameters obtained using the improved grid search with 5-fold cross validation has been proposed to reach the optimal classification performance. Methods: Two data sets, Wisconsin Original Breast Cancer (WOBC) and Wisconsin Diagnostic Breast Cancer (WDBC) data set, were used to classify tumors as benign and malignant. SVM model performs the classification tasks using optimal kernel parameter and penalty parameter using 5-fold cross validation. Discriminant analysis, logistic regression analysis, decision tree, support vector machines were applied to analyze two data sets. Performance of these techniques was compared through accuracy, ROC curves and c-statistics. Results: Our analysis showed that SVMs predicted breast cancer with highest accuracy and c-statistics among four classification models. A comparison of these SVMs indicated that SVM with optimal parameters has much superior performance than SVM with default parameters. Conclusions: Research efforts have reported with increasing confirmation that SVMs have greater accurate diagnosis ability. In this paper, breast cancer diagnosis based on SVM with optimal parameters obtained using the improved grid search with 5-fold cross validation has been proposed. The performance of the method is evaluated using classification accuracy, ROC curves and c-statistics.