类别近似质量约束下的属性约简方法研究

利用近似质量作为度量标准,借助启发式算法求解约简,其本质是根据近似质量的变化情况来找出冗余属性,但这一方法其并未考虑每一个决策类别所对应的下近似集合在约简前后的变化程度.鉴于此,提出了一种基于类别近似质量的属性约简策略,其目标是使得每一个类别的近似质量都满足约简的约束条件.借助邻域粗糙集模型,在UCI数据集上将传统约简策略与类别近似质量约简策略进行了对比分析,实验结果不仅验证了类别近似质量约简策略的有效性,而且表明这种策略依然能够满足传统约简的约束条件.

Attribute reduction constrained by class-specific approximate quality

Based on the measurement of approximate quality,the traditional heuristic algorithm for computing reduction is designed to the find redundant attributes through considering the variation of approximate quality.However,such an approach does not take the variation of lower approximation of each decision class with reduction into account.To fill such a gap,a class-specific approximate-quality-based reduction is proposed.The objective of this strategy is to make the approximate quality of each decision class be acceptable in terms of the constraint of attribute reduction.By using the neighborhood rough set,traditional attribute reduction and class-specific approximate quality based strategies are compared over several UCI data sets.The experimental results tell us that not only the class-specific approximate quality based strategy is effective,but also it satisfies the constraint of traditional attribute reduction.