粗糙集中连续属性离散化的一种新方法

A New Method for Continuous Value Attribute Discretization in Rough Set Theory

提出一种新的间接离散化方法——超曲面法。首先给出了超曲面的定义 ,证明了超曲面为一个高阶的多项式 ,且给出了该多项式的总项数 ;同时证明了超曲面数与最大决策规则数之间的关系。然后提出了应用支持向量机来求取最优超曲面的方法。最后以空军低消耗器材的储存策略为例 ,说明了超曲面的求解过程。实例仿真结果表明 ,用支持向量机来求解超曲面 ,不仅方法简单 ,而且较容易寻得最优解。结果还表明 ,文中提出的超曲面间接离散方法能较好地区分决策表中的决策类别 ,从而获得更为简捷的决策规则。

The discretization of real value attributes is one of the most main problems in rough set theory. This paper presents a new oblique hypersurface discretization method. Firstly, the definition of hypersurface is given and the hypersurface is proved to be a high order polynomial. Item sum of the polynomial is given. At the same time, the relation between the numbers of hypersurface and the maximal numbers of decision rules is also proved. Then a method for obtaining hypersurface by support vector machine is proposed. Finally, taking the store strategy of Air Force low consumption spares as an example, the computational process of the hyperface is illustrated. The example shows that the method for solving the hypersurface by support vector machine is simple and convenient to obtain the optimal solution. It also indicates that the oblique hypersurface discretization method can better distinguish the different categories in decision tables and obtain more simple decision rules.