Skip to Main Content
Core Vector Machine (CVM) is a promising technique for scaling up a binary Support Vector Machine (SVM) to handle large data sets with the utilization of approximate Minimum Enclosing Ball (MEB) algorithm. However, the experimental results in implementation show that there always exists some redundancy in the final core set to determine the final decision function. We propose an approximate MEB algorithm in this paper to decrease the redundant core vectors as much as possible. The simulations on synthetic data sets demonstrate the competitive performances on training time, core vectors' number and training accuracy.