A New Binary Classifier: Clustering-Launched Classification

Abstract

One of the powerful classifiers is Support Vector Machine (SVM), which has been successfully applied to many fields. Despite its remarkable achievement, SVM is time-consuming in many situations where the data distribution is unknown, causing it to spend much time on selecting a suitable kernel and setting parameters. Previous studies proposed understanding the data distribution before classification would assist the classification. In this paper, we exquisitely combined with clustering and classification to develop a novel classifier, Clustering-Launched Classification (CLC), which only needs one parameter. CLC employs clustering to group data to characterize the features of the data and then adopts the one-against-the-rest and nearest-neighbor to find the support vectors. In our experiments, CLC is compared with two well-known SVM tools: LIBSVM and mySVM. The accuracy of CLC is comparable to LIBSVM and mySVM. Furthermore, CLC is insensitive to parameter, while the SVM is sensitive, showing CLC is easier to use.