基于概率密度距离的无监督特征选择方法

Unsupervised feature ranking approach based on probability density interval

在模式识别和数据分析中,经常会遇到数据特征的高维问题。为了有效地进行数据分析,特征维数的削减或特征降维就显得异常重要。针对特征选择这一问题,依据概率密度距离准则,提出一个新的无监督特征排序方法。基于交叉验证的实验结果表明,该方法与现有的方法相比更为有效。

High dimensional datasets often exist in pattern recognition and data analysis, In order to effectively analyze these datasets, reducing their dimensional members is a pivotal step. Based on probability density interval, a novel unsupervised feature ranking approach is proposed, Several cross-validation experimental results demonstrate the advantage of our approach here over others.