An efficient outlier detection method for data streams based on closed frequent patterns by considering anti-monotonic constraints

Abstract

In some practical applications, users pay more attention to a small part of the data that they are interested in, rather than full sets of the data elements, so it is extremely important to improve the security of this small part of the data. For the existing association-based outlier detection methods, they detect the outliers from full set of the data streams, which results in having users to wait for a long period to obtain the detection results. In contrast, most distributions of the outliers are meaningless for the users. To solve this problem, by considering user-specified anti-monotonic constraints, this paper proposes an efficient outlier detection method based on closed frequent patterns for data streams. Also, four deviation indices are designed to accurately calculate the outlier score of each transaction in the sliding window by considering more influencing factors. Then, the top k transactions with the largest outlier score are returned as outliers. Extensive experimental results show that the proposed method can accurately find the outliers from the data streams that satisfy the anti-monotonic constraints with less time cost.