基于邻域粒化和遗传算法的数值型属性约简方法

针对现实中含有数值型属性的决策系统的约简问题,提出了基于邻域粒化和遗传算法的约简方法。该方法采用基于邻域等价关系建立的粗糙集模型,用邻域等价关系度量粗糙集不可分辨关系,通过邻域信息粒子逼近论域空间。构造了遗传约简算法,论述了遗传算法适应度函数的选择,设计了自适应交叉概率,给出了算法的具体实现。对经典数据集和UCI数据库中4个数据库约简的结果证明了算法的有效性和可行性。

Numerical Attribute Reduction Based on Neighborhood Granulation and Genetic Algorithm

In order to reduce the practical decision system including numerical attributes, a reduction algorithm based on neighborhood granulation and genetic algorithm is proposed. In this algorithm, a rough set model is used based on neighborhood equivalence, the indiscernibility relation is measured by neighborhood relation, and the universe spaces is approximated by neighborhood information granules. The choice of fitness function and the adaptive across probability is discussed, and the reduction algorithm is presented as well. Furthermore, the dependency function is used to evaluate the significance of numerical attributes. The validity and feasibility of the algorithm are demonstrated by the results of experiments on a classical data set and four UCI machine learning databases.