Minimal weighted infrequent itemset mining-based outlier detection approach on uncertain data stream
Minimal weighted infrequent itemset mining-based outlier detection approach on uncertain data stream
复制标题
DOI：
10.1007/s00521-018-3876-4
复制
发表时间：
2018-11
影响因子：
6
通讯作者：
Saihua Cai;Ruizhi Sun;Shangbo Hao;Sicong Li;Gang Yuan
中科院分区：
文献类型：
--
作者：
Saihua Cai;Ruizhi Sun;Shangbo Hao;Sicong Li;Gang Yuan
异常值是影响基于数据的预测和其他一些基于数据的处理的准确性的关键因素；因此，必须尽快有效地检测出异常值，以提高数据的可信度。近年来，针对静态数据和精确数据提出了大规模异常值检测方法；然而，之前的工作没有考虑每个项目的不确定性和权重信息。而且，传统的异常值检测方法仅以每个数据元素的偏差程度作为判定异常值的标准；因此，检测到的异常值不符合异常值的定义（即很少出现并且与大多数其他数据不同）。针对这些问题，本文提出一种适用于不确定数据流的基于最小加权非频繁项集挖掘的离群点检测方法MWIFIM-OD-UDS，在考虑数据流特征的同时，有效地检测出频率很少、不确定且项集具有一定权重的隐含离群点。特别是，提出了一种基于矩阵结构的方法，即MWIFIM-UDS，从不确定的数据流中挖掘最小加权非频繁项集（MWiFI），然后，基于挖掘的MWiFI和设计的偏差指标，提出了MWIFIM-OD-UDS方法。实验结果表明，所提出的 MWIFIM-OD-UDS 方法在运行时间和检测精度方面优于基于频繁项集挖掘的异常值检测方法 FindFPOF 和 LFP。
Outliers are a critical factor that affects the accuracy of data-based predictions and some other data-based processing; thus, outliers must be effectively detected as soon as possible to improve the credibility of the data. In recent years, massive outlier detection approaches have been proposed for static data and precise data; however, the uncertainty and weight information of each item was not considered in this prior work. Moreover, traditional outlier detection approaches only take the deviation degree of each data element as the standard for determining outliers; therefore, the detected outliers do not fit the definition of an outlier (i.e., rarely appearing and different from most of the other data). Aimed at these problems, a minimal weighted infrequent itemset mining-based outlier detection approach that can be applied to an uncertain data stream, called MWIFIM–OD–UDS, is proposed in this paper to effectively detect implicit outliers, which have a rarely occurring frequency, uncertainty and a certain weight of the itemset, while the characteristics of the data stream are considered. In particular, a matrix structure-based approach that is called MWIFIM–UDS is proposed to mine the minimal weighted infrequent itemsets (MWiFIs) from an uncertain data stream, and then, the MWIFIM–OD–UDS method is proposed based on the minedMWiFIsand the designed deviation indexes. Experimental results show that the proposed MWIFIM–OD–UDS method outperforms the frequent itemset mining-based outlier detection methods, FindFPOF and LFP, in terms of its runtime and detection accuracy.