An Implementation of Support Vector Machines and Generalized Discriminant Analysis on Iris and Hepatitis Datasets

Abstract

Support Vector Machines (SVM) is a supervised learning method used for classification. The learning strategy of SVM is based on structural risk minimization principle, so SVM has a better ability to generalize than other methods which depend on empirical risk minimization principle. However, when any classification methods face a dataset which is linearity inseparable, they will face a dificulty to classify the dataset. This problem results in low classification rate averages. To anticipate this problem, it is desirable to use Generalized Discriminant Analysis (GDA) as feature extractor. We expect that using GDA will give a better classiication rate averages because it can minimize the distances of data within the same classes and maximize the distance between the different classes. This paper presents a comparison of Support Vector Machines with and without using GDA for Iris and Hepatitis datasets classiication. It is shown that the use of GDA can yield a classiication rate averages of more than 93% for Iris dataset and 95% for Hepatitis dataset.