Support Vector Machine which is invented by Vapnik and Cortes in 1995, belongs to the statistical learning theory. Its application hes been tremendously increased over the years due to its prominent theoretical properties. The basic idea behind the support vector machine is to find the optimal hyper plane for linearly separable data. However, the patterns that are not linearly separable can be transformed from original space into new space by means of the famous kernel function such as linear, polynomial, RBF and etc. The presence of outlying observation can adversely affect the performance of support vector machine and will lead to the subsidence of its accuracy. In this paper, we provide a graphical depiction of data by using the high breakdown robust measure, namely the Mahalanobis distance based on the re-weighted minimum covariance determinant estimator. The so called method, "outlier map" is very popular in the multivariate robust statistics. It can be use to depict the structure of the data with any dimension. Using data from both simulation and real world studies, illustrated that the outlier map based on the robust Mahalanobis distance is the ability to recognize the outlying and misclassified samples in the data.

Keywords