Evidential Editing K-Nearest Neighbor Classifier

摘要

One of the difficulties that arises when using the K-nearest neighbor rule is that each of the labeled training samples is given equal importance in deciding the class of the query pattern to be classified, regardless of their typicality. In this paper, the theory of belief functions is introduced into the K-nearest neighbor rule to develop an evidential editing version of this algorithm. An evidential editing procedure is proposed to reassign the original training samples with new labels represented by an evidential membership structure. With the introduction of the evidential editing procedure, the uncertainty of noisy patterns or samples in overlapping regions can be well characterized. After the evidential editing, a classification procedure is developed to handle the more general situation in which the edited training samples are assigned dependent evidential labels. Two experiments based on synthetic and real data sets were carried out to show the effectiveness of the proposed method.