DB-Outlier Detection Algorithm Using Divide and Conquer Approach over Dynamic DataStream

Anomaly detection is currently an important and active research problem in many fields and involved in numerous applications. Most of the existing methods are based on distance measure which can produce better results as compared to other methods. But in case of datastream these methods are computationally less efficient. In order to find some specified number of neighbors for each data element make these methods inappropriate for datasteam environment, because of the huge volume of DataStream. Moreover, declaring a point as an outlier as it arrives, can often lead us to a wrong decision. In this paper we introduced an algorithm which divide incoming stream in chunks and make preliminary clusters in each chunk during the first stage. In the second stage K-nearest neighbor approach for outlier detection is applied to each cluster. In this way we can reduce the nearest neighbor searches for each data element as well as can detect better outliers with evolution of data stream. Several experiments on different datasets confirm that our technique can find suitable outliers with low computational cost than the other exiting distance based approaches for outlier detection over data stream.