Using a novel support vector machines for efficient classification

摘要

An Improved Support Vector Machines was proposed which starts with a small set and then sequentially expands to include feature space informative data points into the set. These feature space informative data points will be identified by solving a small least squares problem. The approach provides a mechanism to determine the set size automatically and dynamically and the set generated by this method will be more representative than the one by purely random selection. All advantages of SVM for dealing with nonlinear classification problem are retained.