基于遗传算法和变精度粗糙集的离散化算法

目前常用的离散算法多为单属性离散化算法.利用该类算法对多维连续属性进行离散化时，逐次对单个属性进行离散化，割裂了多维属性之间的关系.基于此提出了一种基于遗传算法和变精度粗糙集的多属性离散化算法.该算法基于变精度粗糙集所具有的较好数据分类容错和抗噪能力，通过变精度粗糙集近似分类精度建立遗传算法适宜度评价函数，并利用遗传算法在多维连续属性候选断点集上寻找最优断点子集.基于UCI数据集比较了所提算法与多种常用的离散化算法的差异，实验结果表明，该算法可以获得相对较好的离散化效果.

Discretization algorithm based on genetic algorithm and variable precision rough set

The commonly used discretization methods are suitable for the discretization of single attribute. They successively discretize each attribute while used for the discretization task of multiattributes. This kind of discretization methods does not take into the relations between attributes. A continuous multiattributes discretization method based on genetic algorithm and variable precision rough set was proposed. The method used genetic algorithm to find the global optimal cut points of continuous multiattributes. And the fitness function was constructed based on the accuracy of approximation classification of variable precision rough set. To validate the effectiveness of the proposed discretization method， it was applied to some UCI data sets. And the experimental results indicate that the proposed discretization method can acquire relative better discretization results.