OUTLIER DETECTION FOR DYNAMIC DATA STREAMS USING WEIGHTED K-MEANS

摘要

This paper presents a new k-means type clustering algorithm that can calculate weights to the variables. This method is efficient for dynamic data streams in order to overcome the global optimum problems. The variable weights produced by the algorithm measures the importance of variable in clustering and can be used in variable selection in which the data items with similar properties are grouped into clusters, the new approach of applying this weighted k-means on dynamic data streams is carried out in order to have efficient outlier detection within the user specific threshold value.