An efficient approach for outlier detection from uncertain data streams based on maximal frequent patterns

•We propose MFP-OD for outlier detection from uncertain data streams.•We pay more attention to the associations between each data instance.•We study the deviation index of each transaction and give the interpretation.•Our method outperforms five baseline methods on four datasets. Outlier identification is an important technology to improve the credibility of data and aims at detecting patterns that rarely appear and exhibit a significant difference from other data. However, the detection accuracy achieved by the simple deviation factors of existing pattern-based outlier detection methods is not competitive. In addition, given the large scale of uncertain data streams, the efficiency of many pattern-based outlier detection methods is not high because they use a vast number of frequent patterns to conduct the outlier detection. In this paper, to contend with the uncertain data streams, we propose a maximal-frequent-pattern-based outlier detection method, namely, MFP-OD, for identifying the outliers with a lower time cost. For further improving the detection accuracy of existing outlier detection methods, we design three deviation factors to measure the deviation degree of each transaction. The experimental results indicate that the proposed MFP-OD method can quickly and accurately identify the outliers from uncertain data streams.