基于ICA的异常数据挖掘算法研究

Study of Outlier Data Mining Algorithm Based on ICA

在传统的独立成分分析方法中,没有考虑异常数据值对分离性能的影响。该文提出了一种基于影响函数的检测方法,通过该方法可以发现隐藏在观测数据中的异常成分。利用影响函数对数据进行投影分析,对混入脉冲噪声的观测信号进行盲源分离,从而实现对脉冲噪声的消除。实验仿真结果表明,该方法可以有效且可靠地检测出所观察信号中的异常数据。

In the traditional study of independent component analysis （ICA）, the outlier data had not been considered. This paper proposes a method based on influence function to find the outliers from the observed data in ICA. General, outliers have a significant influence on the separation performance of ICA. Using the influence functions to project the observed data, the impulsive noisy components which mixed in the observed data can be eliminated from the normal data. The experimental results demonstrate the effectiveness of proposed method.