An improvement in support vector machine classification model using grey relational analysis for cancer diagnosis

Abstract

To further improve the accuracy of classifier for cancer diagnosis, a hybrid model called GRA-SVM which comprises Support Vector Machine classifier and filter feature selection Grey Relational Analysis is proposed and tested against Wisconsin Breast Cancer Dataset (WBCD) and BUPA Disorder Dataset. The performance of GRA-SVM is compared to SVM’s in terms of accuracy, sensitivity, specificity and Area under Curve (AUC). The experimental results reveal that GRA-SVM improves the SVM accuracy of about 0.48 by using only two features for the WBCD dataset. For BUPA dataset, GRA-SVM improves the SVM accuracy of about 0.97 by using four features. Besides improving the accuracy performance, GRA-SVM also produces a ranking scheme that provides information about the priority of each feature. Therefore, based on the benefits gained, GRA-SVM is recommended as a new approach to obtain a better and more accurate result for cancer diagnosis.