关键词: 孤立点检测, 距离, DBSCAN算法, 剪枝

Abstract: Most of the existing outlier detection algorithms need to preset the number of outliers, and also lack of detection capability of non-uniform data set. In view of the above problems, it puts forward the two-part outlier detection algorithm based on clustering, this algorithm first uses DBSCAN clustering algorithm to produce suspected outlier set, then pruning strategy is used for pruning data set, and the outlier detection algorithm based on improved distance is used to produce the sorting set of the points which most likely to be outliers. Eventually the isolated point set is determined by the intersection of the two sets. The algorithm doesn’t need to preset the number of outliers, with the higher accuracy and detection efficiency, and is not sensitive to the distribution of the data set. The experimental results on data set show that the algorithm can effectively and accurately identify the outliers.

Key words: outlier detection, distance, DBSCAN algorithm, pruning