论文部分内容阅读
Pawlak提出的基于属性重要度的约简算法是常用的算法之一,它通过计算等价关系对论域划分的粒度来度量属性的重要度。但用该算法计算每一个属性的重要度时,都要计算不同等价关系对整个论域的划分,计算复杂度非常高。受决策树划分子集思想的启发,对基于属性重要度的属性约简算法进行了改进,提出了一种基于划分子集的属性约简算法。在核属性集形成划分的基础上,通过在核属性中添加非核属性从而形成更细的划分,如此反复。在保持正域不变的框架下,形成最细化分的属性集就是一个约简。理论分析显示该算法减少了求属性约简的计算时间复杂度,提高了求属性约简的效率。
Pawlak’s reduction algorithm based on attribute importance is one of the most commonly used algorithms. It measures the importance of attributes by computing the granularity of equivalence relations. However, when using this algorithm to calculate the importance of each attribute, it is necessary to calculate the division of the entire universe by different equivalence relations, and the computational complexity is very high. Inspired by the thought of decision tree partitioning subsets, an attribute reduction algorithm based on attribute importance is improved, and a attribute reduction algorithm based on subset partitioning is proposed. Based on the division of nuclear attribute set, a finer division is formed by adding non-nuclear attributes to the nuclear attribute, which is repeated. Under the framework of keeping the positive domain unchanged, the attribute set that forms the most detailed score is a reduction. Theoretical analysis shows that this algorithm reduces the computational time complexity of attribute reduction and improves the efficiency of attribute reduction.