混合决策系统启发式增量属性约简方法研究

对于混合决策系统的属性约简,现有方法主要存在动态效果不佳、复杂度过高,以及约简精度差等问题,为此,提出一种启发式增量属性约简方法。针对混合决策系统的动态波动,基于粗糙集建立了邻域关系模型,根据邻域相对差异对增量属性进行更新。同时,为进一步增强约简算法的动态适应性,引入条件熵求解相对差异。考虑到单纯利用邻域依赖虽然有利于处理样本的分布不均,但是很难获得良好的属性评估,引入粒度模型进行优化,将邻域关系采用粒度重新描述,从而细化邻域关系。利用邻域依赖性得到决策属性度量,构造启发计算,同时,通过条件和决策间的关联度,以及粒度模型的单调,求解出条件和决策共同约束下的邻域关系。再根据决策属性度量作为启发,直至单一属性对子集决策性能不再有影响,完成属性约简。基于数据集的仿真,验证了提出的启发式增量属性约简方法能够降低约简冗余度和约简长度,有效提高属性约简精度和约简时间效率。

Research on Heuristic Incremental Attribute Reduction Method of Hybrid Decision System

For attribute reduction of hybrid decision system,the existing methods mainly have the problems of poor dynamic effect,high complexity and poor reduction accuracy,Therefore,a heuristic incremental attribute reduction method is proposed.In view of the dynamic fluctuation of the hybrid decision system,a neighborhood relation model was established based on the rough set,and the incremental attribute was updated according to the relative difference of the neighborhood.At the same time,in order to further enhance the dynamic adaptability of reduction algorithm,condition entropy was introduced to solve the relative difference.Considering that the use of neighborhood dependency alone is conducive to dealing with the uneven distribution of samples,but it is difficult to obtain a good attribute evaluation,the particle size model was introduced to optimize,The neighborhood relationship was re described by granularity,so as to refine the neighborhood relationship.Neighborhood dependence was used to get decision attribute measure and construct heuristic calculation.At the same time,through the correlation between conditions and decisions,and the monotony of the granularity model,the neighborhood relationship under the common constraints of conditions and decisions was solved.Then Taking the decision attribute measurement as a heuristic,the attribute reduction was completed until a single attribute no longer affects the decision-making performance of the subset.Based on UCI data set,the simulation experiment verifies that the proposed heuristic incremental attribute reduction method can reduce the redundancy and reduction length and effectively improve the accuracy and time efficiency of attribute reduction.