Data augmentation for linearly separable feature space
EunSang Bak · 2005
Most of the classification methods involve data reduction procedure before making decisions by given classifiers. In fact, selection of classifier is directly related with data reduction procedure. In this paper, a new method called data augmentation is proposed to increase separability in feature space. As opposed to the other methods, it increases the dimension of the feature vectors and makes a corresponding feature space linearly separable. In addition, it involves a discriminant function as its classifier and shows better classification results in the feature space after data augmentation. The details of data augmentation in the classification method are mathematically derived and its performances are evaluated by practical applications.