基于粒度计算的特征选择方法

从粒度计算的划分模型出发,重新定义了相容决策表的约简,并给出了一种新的基于粒度计算的属性约简算法.该算法以信息熵作为启发信息,通过逐渐增加属性构成条件属性集相对于决策属性的约简,再通过删除约简中的所有不必要属性,得到最小约简.该算法有效地降低了计算属性约简的时间复杂度,可以用于较大规模数据集的特征选择.在5个公开的基因表达数据集上的实验证明了该算法能找到高区分能力的特征子集.

Method for Feature Selection Based on Granular Computing

The reduction of consistent decision table is redefined from the point of view of partition module of granular computing, and a novel algorithm to find an optimal reduction with low time complexity is proposed. Reduction is constructed by adding attributes using information entropy as the heuristic information, superfluous attributes are deleted from the reduction to get a minimal reduction. The experiment results on five public gene expression profiles demonstrate the usefulness of this new method for feat...