A Study on Supervised Classification Approach Using Support Vector Machines

Abstract

Support Vector Machine (SVM) is the key for all the classification problems. It is a pow- erful tool that finds a classifier to separate two different classes that maximizes the margin between them. It was developed for binary classification and further it was used for solv- ing multi-class problems. In this dissertation we have reviewed the study of support vector machines for binary and multi-class classification problem. This work is divided into the following four chapters. First chapter consists of the introduction to Machine learning algorithm. We have also included the history, importance and some applications of machine learning. Some important definitions that are required throughout the dissertation and some models that were used earlier are also written in this chapter. In second chapter the basics of Support Vector Machines (SVM) is introduced in this chapter. This involves some derivations for linear as well as non-linear classification. Further, algorithms for solving such problems are also presented in this chapter. In third chapter we have reviewed the work of O.L. Mangasarian and David R. Musicant, Lagrangian Support Vector Machines (LSVM). Their work consists of iterative method for binary classification unlike Support Vector Machines (SVM) which uses quadratic programming for finding support vectors. Algorithms for both linear and non-linear classifi- cation is explained in this chapter. Further, comparison between SVM and LSVM on various data sets is also included. In the last chapter we have explained Lagrangian Support Vector Machines for multi-class problems by J. P. Hwang, B. Choi, I. W. Hong, and E. Kim. Firstly, SVM for multi-class problems is described. Further, a method to solve multi-class problems (LSVMMPAC) con- sidering all the classes at once is presented in this chapter. Later, LSVMMPAC for non-linear problems using kernel methods is explained in this chapter.