Multi-class Classifier based on Support Vector Machine with Application to Ordinal Data

Abstract

Support vector machine initially developed to perform binary classification. This paper presents a multi-class support vector machine classifier and ordinal regression to classify the type of bone mineral density. This paper compares the performance of four multi-class approaches, one-against-all, one-against-one, Weston and Watkins, and Crammer and Singer. Results from our real life data conclude that Crammer and Singer may be better approach depending on training error and the percentage of correctly classified test data. Also, we fined that the training error become more less when the regulization parameter C and kernel parameter  become large.