Study on support vector machine as a classifier

Abstract

SVM [1], [2] is a learning method which learns by considering data points to be in space. We
studied different types of Support Vector Machine (SVM). We also observed their
classification process. We conducted10-fold testing experiments on LSSVM [7], [8] (Least
square Support Vector Machine) and PSVM [9] (Proximal Support Vector Machine) using
standard sets of data. Finally we proposed a new algorithm NPSVM (Non-Parallel Support
Vector Machine) which is reformulated from NPPC [12], [13] (Non-Parallel Plane
Classifier). We have observed that the cost function of NPPC is affected by the additional
constraint for Euclidean distance classification. So we implicitly normalized the weight
vectors instead of the additional constraint. As a result we could generate a very good cost
function. The computational complexity of NPSVM for both linear and non-linear kernel is
evaluated. The results of 10-fold test using standard data sets of NPSVM are compared with
the LSSVM and PSVM.