Attribute reduction method based on approximate quality and conditional entropy

Abstract

The invention discloses an attribute reduction method based on approximate quality and conditional entropy. The method comprises: using a heuristic algorithm for solving approximate quality reduction;wherein the input is a neighborhood decision system DS = (U, ATU {d}), constraint conditions, metrology criterion gamma, the output is an approximate mass reduction red. Although an attribute reduction result based on a single criterion on a neighborhood rough set can meet constraint conditions of corresponding measurement indexes, the attribute reduction result can meet the constraint conditions. However, other measurement criteria cannot be still met. Therefore, the algorithm provided by the invention fuses multiple criteria of approximate quality and conditional entropy to serve as a new reduction method of measurement indexes, and experimental results show that the new reduction not only can reduce the conditional entropy on the basis of keeping the approximate quality unchanged obviously, but also can effectively improve the classification precision.

Description

technical field [0001] The invention relates to an attribute reduction method based on approximate quality and conditional entropy. Background technique [0002] Rough set is a mathematical tool proposed by Polish scholar Pawlak to describe imprecision and uncertainty. Neighborhood rough set is an important extended model of classical rough set. Compared with traditional rough set, neighborhood rough set not only It can be applied to deal with continuous numerical values, and due to the existence of the neighborhood radius, a multi-granularity structural framework can be naturally formed according to the radius of different sizes. Complex data has many advantages such as strong adaptability and easy realization of incremental calculations, which have attracted extensive attention from many scholars. [0003] Similar to the research of other extended rough set models, attribute reduction also occupies a core position in the research of neighborhood rough sets. Due to the dif...

AI Technical Summary

Problems solved by technology

[0003] Similar to the research of other extended rough set models, attribute reduction also occupies a core position in the research of neighborhood rough sets. Due to the difference in demand goals or extended models, many metrics can be used to define attribute reduction, such as : Li et al. studied attribute reduction under category approximation quality; from the perspective of classification learning, in order to improve the classification performance of the algorithm after reduction, Hu et al. proposed the concept of neighborhood decision error rate; Zhang et al. The rough set model introduces information entropy, and information entropy can not only be used as a tool to describe uncertainty, but also can reflect the pros and cons of classification performance to a certain extent. The traditional method of obtaining reduction usually uses a heuristic framework A single index is selected as the measurement criterion. Although the concept of attribute reduction under a single measurement criterion has the advantages of clear goals and easy promotion, in practical applications, comparing the results of attribute reduction under different criteria, it is not difficult to find that due to the selected measurement criterion If it is too single, the following problems will occur:

[0004] 1) It is difficult for the obtained reduction to satisfy the constraints of multiple different criteria at the same time. For example, the attribute reduction based on the approximate quality simply satisfies the constraints of the approximate quality, but there is no guarantee that the reduction result can still satisfy the Constraints on conditional entropy or neighborhood decision error rate;

[0005] 2) The reduction obtained does not guarantee a greater advantage in classification performance,

Images