基于遗传算法和区分矩阵的属性约简

在保持知识库分类能力不变的条件下,为了获得决策系统中属性的最小约简,本文利用遗传算法在全局寻优、避免算法陷入局部收敛方面的优势,结合区分矩阵能够很容易地计算出属性约简和核的良好性质,提出了一种基于遗传算法和区分矩阵的属性约简算法。该算法将区分矩阵嵌入遗传算法中,充分发挥各自在求解最小属性约简方面的优良特性,其中遗传算法主要流程基本不变,适应值函数选取引入区分矩阵的性质。实验结果表明,在可接受的时间内得到了最小约简,约简结果优于HU算法和传统遗传算法,达到了预期的结果。

An Attribute Reduction Algorithm Based on Genetic Algorithms and Discemible Matrixes

In order to obtain the relatively minimal reduction of the attributes in a decision making system without changing the classification capacity of the knowledge base, to take advantage of genetic algorithms in global optimization while avoiding the algorithm runing into local convergence, and to combine the property of discemible matrixes by calculating of the reduction of properties and the core easily, an attribue reduction algorithm based on genetic algorithms and discemible matrixes is proposed. The algorithm embeds the dicemible matrix into the genetic algorithm, and develops each of the good properties to its greatest extent in working out the relatively minimal reduction, in which the main process of GA is not changed, and the property of the discemible matrix is adopted in selecting the fitness function. The experimental results show that the algorithm can achieve the desired minimal reduction results in an acceptable time,which is better than the ones obtained from the HU algorithm and the traditional genetic algorithms.