摘要
相比于经典粗糙集方法,模糊粗糙集方法避免了数据离散化的过程,减少了信息损失。但基于传统模糊粗糙集的属性约简并未考虑实际应用中数据的测试代价,为解决这一问题,提出了一种近似质量与测试代价相融合的适应度函数,并利用遗传算法以求得具有较小测试代价的约简。最后,采用UCI中的8组数据集对基于新适应度函数的遗传算法与经典的启发式算法进行对比分析,实验结果表明,遗传算法相较于启发式算法能够在保证近似质量不发生明显变化的情况下获得具有较低测试代价的约简。
Compared with the classical rough set approach, fuzzy rough set approach avoids the process of data discretization and then decreases the loss of information. However, the attribute reduction based on traditional fuzzy rough set does not consider the test cost of data in many practical applications. To solve such problem, a fitness function which fuses both approximate quality and test cost is proposed, the genetic algorithm is then employed to find reduct with smaller test cost. Finally, the comparison between genetic algorithm based on new fitness function and classical heuristic algorithm is tested on eight UCI data sets. The experimental results tell us that by comparing heuristic algorithm, genetic algorithm can achieve reduct with lower test cost without the obvious changing of approximate quality.
出处
《电子设计工程》
2015年第11期5-7,11,共4页
Electronic Design Engineering
基金
国家自然科学基金(61100116)
江苏省自然科学基金(BK2011492)
江苏省高校自然科学基金(13KJB520003
13KJD520008)
关键词
属性约简
决策系统
模糊粗糙集
测试代价
attributes reduction
decision system
fuzzy rough set
test cost
