Effects Of Outliers In Support Vector Machines

Support Vector Machines (SVM) is a new technique of classification that has received much attention in recent years. In many applications, the SVM has shown better performance than machine learning methods, and it has been introduced as a powerful tool for solving classification problems. The SVM was originally developed for binary classification, but was later generalized to problems with various classes using different approaches. Currently, the SVM can be applied to an extensive list of scientific and real life problems. This thesis describes the SVM method for the separable case, the non-separable case, and for multiple classes. Several outlier detection methods are discussed including the support vector description method as well as the use of SVM for one class classification to detect outliers. Experiments with the SVM classifier were made and it is empirically shown that after the elimination of the detected outliers the misclassification error rate of the SVM classifier is improved. All experiments were carried out on 5 data sets available at the Machine Learning Database Repository of the University California, Irvine.