A Novel Supervised Learning Based on Density

– Because prototype based classifiers are both easy and reasonable methods, there have been many studies on similarity based supervised learning. In order to detect each class region, they should not only appropriately locate the prototypes, but also deal with overfitting and instability. In this study, by considering all these criteria, we develop a new classifier method based on the prototypes selected from dense patterns. While the method determines details of the prototypes, it evades overfitting according to relation of the correct classification accuracy and the number of prototypes. Because of its similarity in point of architecture, we compare it with learning vector quantization (LVQ) method by using some synthetic and benchmark datasets. This comparison shows that our method is better than the other, and it may cause new suggestions on classification and some real applications