高维数据挖掘中基于中位数回归的特征提取新方法

摘要

为降低噪声对数据特征提取(变量选择)效果的不利影响,基于中位数回归分析方法,利用变量选择降维技术(正则化估计),提出了一种稳健、有效的特征提取(变量选择)新方法,并具体给出了估计算法,该算法具有快速计算的特点.实验结果表明,新方法能够有效地对高维数据集进行估计和变量选择,且具有较高的准确性,即使数据中的信噪比很低时,该方法仍具有较好的效果.因此,该方法为高维数据挖掘特征提取提供了稳健且有效的方法.%In order to reduce the unfavorable influence of nosie existing to the feature extraction(variable selection) , this paper proposed a robust and effictive method based on median regression and dimensional reduction technique of variable selection (regularized estimation). Moreover, it gave the detail of algorithm which possessed the advantage of fast computation. Simulations show that the new method can effictively estimate and select important variables with high correctness. Even in the case of very low signal to noise ratio, this method still has good result compared with other method. It is really an efficient and robust feature extraction method for high-dimensional data mining.