A Kernel-based Matrixzed One-Class Support Vector Machine
International Journal of Hybrid Information Technology
One-class support vector machine (OCSVM) is an important and efficient classifier used when only one class of data is available while others are too expensive or difficult to collect. It uses vector as input data, and trains a linear or nonlinear decision function in vector space. However, the traditional vector-based classifiers may fail when input is matrix. Therefore, it makes sense to study matrixzed classifiers which can make use of the structural information presented in the data. In this
... paper we propose a matrix-based one-class classification algorithm named Kernel-based Matrixzed One-class Support Vector Machine (KMatOCSVM). It aims to convert the OCSVM to suit for matrix representation data and to deal with nonlinear one-class classification problems. The efficiency and validity of the proposed method is illustrated by four real-world matrixbased human face datasets. In this paper, we propose a nonlinear one-class classifier which directly takes matrix as input. The new algorithm is intended to solve nonlinear classification problems which cannot be dealt with by MatOCSVM. It is named Kernel-based Matrixzed One-Class Support Vector Machine (KMatOCSVM). KMatOCSVM, directly using matrix as input, can help to retain the data topology efficiently in comparison with vector-based classifier. The efficiency and validity of the proposed method is illustrated by real-world matrixbased human face datasets. The rest of this paper is organized as follows. A brief summary of some relevant concepts on the standard OCSVM and MatOCSVM are presented in Section 2. The proposed KMatOCSVM is described in Section 3 and the experimental evaluation is presented in Section 4. Finally, conclusions of our work are drawn in Section5. i are support vectors. Matrixized One-Class Support Vector Machine MatOCSVM is a matrix-pattern-oriented version of OCSVM and can directly classify the samples represented by matrix. Given a set of training samples , 1 X i il , each of the training sample Generally speaking, for nonlinear classification problems, kernel-based classifiers often take longer training time to achieve higher accuracy than those of linear classifiers. In contrast, the proposed KMatOCSVM takes shorter training time than that of MatOCSVM based on the experimental results of 3 out of 4 datasets. A possible explanation is because of the iteration method corresponding to KMatOCSVM and MatOCSVM, which leads to the fact that the training time depends on the number of iterations. Therefore, the time cost on calculating the kernel matrix can be negligible. In summary, the proposed kernel-based KMatOCSVM has an outstanding performance compared to the linear MatOCSVM, both on the AUC and the training time.