基于决策表相容度和属性重要度的连续属性离散化算法  

Continuous Attribute Discretization Algorithm Based on Decision Table Compatibility and Attribute Importance

作　　者：王成宇 林名驰[1] WANG Chengyu;LIN Mingchi(Department of Management Engineering and Equipment Economy,Naval University of Engineering,Wuhan 430033;Studio of Engineering Management,No.92690 Troops of PLA,Sanya 572000)

机构地区：[1]海军工程大学管理工程与装备经济系,武汉430033 [2]92690部队施工管理室,三亚572000

出　　处：《舰船电子工程》2022年第4期43-48,共6页Ship Electronic Engineering

摘　　要：基于粗糙集理论的模型筛选与组合预测方法对于舰船维修费用的预测具有较强的应用价值,然而在连续属性离散化方面却存在部分问题。针对由多种预测模型预测值构成的数据表的离散化及模型筛选问题,分别分析了有监督离散化算法和无监督离散化算法的局限性与适用性,并针对无监督离散化算法可能改变原有不可分辨关系、已有的改进算法可能存在的断点冗余以及分别考虑决策表与条件属性的不相容度导致的计算复杂的问题,引入决策表相容度作为反馈信息,从整体上考虑决策表的相容度,初次离散化选取数值合理的断点数,并结合各条件属性的属性重要度对各条件属性进行排序,通过逐次对决策表相容度进行判别,依排序情况逐个对条件属性的断点数进行调整,以达到离散化效果并保证决策表的相容度不变。通过例证分析,验证了该算法的有效性。Model selection and combination forecasting methods based on rough set theory has strong application value for the forecasting of ship maintenance costs,but there are some problems in the discretization of continuous attributes.Aiming at the dis⁃cretization and model selection of data tables composed of the forecasting values of multiple forecasting models,the limitations and applicability of the supervised discretization algorithm and the unsupervised discretization algorithm are analyzed respectively,and the unsupervised discretization algorithm may be changed.The original indistinguishable relationship,the possible breakpoint re⁃dundancy of the existing improved algorithm,and the computational complexity caused by the incompatibility between the decision table and the condition attributes are considered separately.The decision table compatibility is introduced as feedback information,the compatibility of the decision table is considered from the overall,a reasonable number of breakpoints are selected for the first discretization,the attribute importance of each condition attribute are combined to sort the condition attributes,and the compatibili⁃ty of the decision table is judged one by one according to the sorting situation.The number of breakpoints of conditional attributes are adjusted to achieve the discretization effect and ensure the consistency of the decision table remains unchanged.Through the ex⁃ample analysis,the effectiveness of the algorithm is verified.

关 键 词：离散化 决策表相容度 属性重要度 连续属性 

分 类 号：O212.1[理学—概率论与数理统计]