基于时序离群检测的新的分段方法

New segment method of temporal data for outlier detection

在对时序数据进行离群检测之前,一般先将原时序数据划分为若干个子序列,以便降低计算复杂度。现有的子序列划分方法一般是依据应用要求进行,而在某些情况下应用要求无法转换为有效的子序列划分方法。因此,提出从时序数据自身特点出发,得到突变系数和重要点,依据重要点和突变系数的新的划分方法,并以微软的股票数据进行测试。实验结果表明,分段方法不依赖于应用要求,具有简单、直观的特点,与相关算法相比,具有更高的检测精度。

General approaches for outlier detection need to divide temporal data into sub-sequences so as to reduce complexity. The existing methods divide temporal data by application, which is not available on some occasions. A new segment method based on the properties of temporal data is proposed, which divided temporal data by combining important point with their breaking factor （BF）. Microsoft stock price series are used for testing. The results show that the segment method is simple, intuitive, independent of application, and outperforms relevant method.