测试代价敏感的粗糙集方法

• 作 者: 鞠恒荣 周献中 杨佩 李华雄 杨习贝
• 机 构: 南京大学工程管理学院,江苏科技大学计算机科学与工程学院,南京大学智能装备新技术研究中心,南京大学软件新技术国家重点实验室,
• 出 处: 系统工程理论与实践
• 发表日期: 2017年01期
• 关键词: 粗糙集模型 属性约简 测试代价敏感 遗传算法 启发式算法 属性集合 下近似集 正域 不确定性 属性子集

在粗糙集模型中,α量化不可分辨关系是强与弱不可分辨关系的推广形式.然而值得注意的是,基于这三种不可分辨关系的粗糙集并未考虑数据中属性的测试代价.为解决这一问题,提出了测试代价敏感的α量化粗糙集模型,从二元关系的角度使得粗糙集模型代价敏感,并将新模型与基于强不可分辨、弱不可分辨以及传统α量化不可分辨关系的粗糙集模型进行了对比分析.进一步地,通过分析传统启发式算法在求解约简的过程中未考虑降低代价这一不足之处,提出一种新的属性适应性函数,并将其应用于基于遗传算法的约简求解中.实验结果表明该方法不仅可以降低由边界域所带来的不确定性而且同时降低了约简后的测试代价.

In the rough set model, the α-quantification inextensible relation is the generalized form of strong and weak unresolvable relations, but it is worth noting that the rough set based on these three irreducible relations does not consider the test cost of the attributes in the data. In this problem, a cost-sensitive α-quantized rough set model is proposed, which makes the cost of the rough set model sensitive from the perspective of binary relations. Moreover, the new model is compared with the inequality based on the strong inseparability, the weak indissolubility and the traditional α quantification Rough set models are compared and analyzed.Furthermore, a new attribute adaptive function is proposed based on the analysis of the traditional heuristic algorithm without reducing the cost in the process of solving the reduction, In genetic algorithm reduction, the experimental results show that this method can not only reduce the uncertainty brought by the boundary region, but also reduce the test cost after reduction.