| Support Vector Machines (SVMs) are learning systems for solving classification and regression problems and considered one of the most powerful tools in machine learning. SVMs utilize kernel methods that provide a framework for solving nonlinear classification or regression problems in a higher dimensional space. Finding the SVM solution can be regarded as estimating the center of the largest hypersphere that can be inscribed in the set of consistent hypotheses called a version space. In this research, we utilize one of several possible centers of the version space that can improve the generalization performance, the so called p-Center. The concepts of the p-Center of a convex polytope and version space are presented. Using the p-Center approach, we develop the p-Center based kernel machines for regression analysis and multiclass classification. Then, an algorithm for performing active learning with the p-Center machine is proposed. The objective of active learning is to select the instances to be labeled and included in the training set. Applications of the proposed algorithms are also investigated. The experimental results show that the performance of the proposed p-Center based kernel machines is competitive compared to SVMs. |
