基于ICA与MViSOM的孤立点挖掘模型

Outlier Mining Model Based on ICA and MViSOM

本文提出了一种基于独立成分分析(ICA)与改进的可视化诱导自组织映射(MViSOM)的孤立点挖掘模型——IMVOM模型,该模型用ICA方法对观测到的多维随机向量进行独立成分分解,得到一个独立成分数据集,然后用改进的MViSOM方法取得数据的可视化。该模型充分结合“人类擅长于模式识别的能力”与“电脑擅长于大量地记忆、快速地计算的能力”的双方优点进行孤立点的挖掘,避免了对高维数据内部结构的复杂探测,从而克服了高维数据集孤立点挖掘过程中的一些困难。实验结果也验证了所提模型的合理性。

IMVOM,Outlier Mining Model Based on ICA ＆ MViSOM, is presented in this paper. This model firstly transforms an observed multidimensional random vector into mutually independent components by ICA, and then achieves visibility of high-dimensional data by MViSOM. Combined the pattern recognition capacity of human being with the calculating capacity of computer, this model can finish mining the outliers by avoiding of detecting the complex inner structure of data and overcoming some difficulties of outlier mining of high-dimensional data. In the end, the proposed model＇s correctness and reasonableness are also validated by the experiment results in this paper.