基于奇异值分解的异常切片挖掘

切片操作是联机分析处理的主要功能之一,在决策支持应用中发挥着重要作用.由于人工的切片过程非常低效,且易忽略重要信息,提出了一种自动、智能的异常切片挖掘方法.该方法基于奇异值分解技术来提取切片的数据分布特征,然后在提取出的奇异值特征之上,利用基于距离的孤立点检测技术发现异常的切片.在人工生成的数据和实际应用的切片数据上所作的实验结果都表明了该方法的高效性和可行性.

Exceptional Slices Mining Based on Singular Value Decomposition

Slice is one of the major operations in on-line analysis processing, which has played an important role in the application of decision support. In this paper, a method of mining exceptional slices is presented for extracting the distribution feature of the slice data based on the technique of the singular value decomposition, and the exceptional slices can be found by utilizing the distance-based outlier detection technique on the singular value feature. The effectiveness of the approach is experimentally demonstrated on the artificial data and the real slices data.