关键词: 聚类, 边界点, k-近邻距离, k-离群度, 边界因子

Abstract: Border is a useful model.In order to detect the boundaries of various shapes effectively，based on the uneven density around the border point，this paper proposes a boundary detection algorithm—BDKD.This algorithm defines the ratio of k-nearest neighbor distance of data objects to the average neighborhood of them as their k-outlier degrees.They are identified as boundary points when the k-outlier of them exceeds the threshold.The results show that BDKD algorithm can accurately detect the boundaries of various clusters and remove the noises.In particular，BDKD algorithm is suitable for the data set of uniform density satisfactorily.

Key words: clustering, border point, k-nearest neighbor distance, k-outlier degrees, border factor