关键词: 数据挖掘, 聚类结果, 测度函数, 自适应性, OD_KC方法

Abstract: Because the behavior pattern of users are always diverse and unpredictability, the traditional outlier detection methods using normal or abnormal models former setting become a difficult problem. To solve this problem, this paper presents a self-adapting outlier detection method based on k-means clustering, called OD_KC algorithm. Based on the unlabeled training samples are clustered by k-means method by different clustering parameter, a measure function is constructed to measure the performance of clustering process to obtain the optimal clustering results, and the small size classes after clustering are took as the outlier model. The outlier detection method based on k-means clustering has the autonomy and adaptability. Specially, the good results also can be obtained when the training data distribution is difficult by the OD_KC method and it has good outlier detection ability. Simulation results on standard datasets demonstrate that excellent detection results can be obtained by this method.

Key words: DM, clustering results, measure function, adaptation, OD_KC algorithm